Datasets:

ppxscal

/

academic_embeddings_cosimrank_test_bfs

like 0

Cloud based centralized task control for human domain multi-robot operations. With an increasing number of assistive robots operating in human domains, research efforts are being made to design control systems that optimize the efficiency of multi-robot operations. As part of the EU funded RoboEarth project, this paper discusses the design of such a system, where a variety of existing components are selected and combined into one cohesive control architecture. The architecture’s main design principle stems from Radestock’s ‘separation of concerns’, which dictates the separation of software architectures into four disjunct components; coordination, configuration, communication and computation. For the system’s coordinating component a Golog based planning layer is integrated with a custom made execution module. Here, the planning layer selects and parametrizes abstract action plans, where the execution layer subsequently grounds and executes the involved actions. Plans and plan related context are represented in the OWL-DL logics representation, which allows engineers to model plans and their context using first-order logic principles and accompanying design tools. The communication component is established through the RoboEarth Cloud Engine, enabling global system accessibility, secure data transmissions and the deployment of heavy computations in a Cloud based computing environment. We desire these computations, such as kinematics, motion planning and perception, to all run on the Cloud Engine, allowing robots to remain lightweight, the instant sharing of data between robots and other algorithms and most importantly, the reuse of these algorithms for a variety of multi-robot operations. A first design of the system has been implemented and evaluated for its strengths and weaknesses through a basic, but fundamental real-world experiment.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

3D Human pose estimation: A review of the literature and analysis of covariates.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Efficient data retrieval for large-scale smart city applications through applied Bayesian inference Recent years have witnessed the proliferation of worldwide efforts towards developing technologies for enabling smart cities, to improve the quality of life for citizens. These smart city solutions are typically deployed across large spatial regions over long time scales, generating massive volumes of data. An efficient way of data retrieval is thus required, for post-processing of the data - such as for analytical and visualization purposes. In this paper, we propose a data prefetching and caching algorithm based on Bayesian inference, for the retrieval of data in large-scale smart city applications. A brute-force approach is used to determine the optimal weight correction factor in the proposed prefetching algorithm. We evaluate the optimized Bayesian prefetching algorithm against the Naïve and Random prefetch baselines, using both simulated and actual data usage patterns. Results show that the Bayesian approach can achieve up to 48.4% reductions in actual user-perceived application delays during data retrieval.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Conformant planning for domains with constraints: a new approach The paper presents a pair of new conformant planners, CPApc and CPAph, based on recent developments in theory of action and change. As an input the planners take a domain description D in action language AL which allows state constraints (non-stratified axioms), together with a set of CNF formulae describing the initial state, and a set of literals representing the goal. We propose two approximations of the transition diagram T defined by D. Both approximations are deterministic transition functions and can be computed efficiently. Moreover they are sound (and sometimes complete) with respect to T. In its search for a plan, an approximation based planner analyses paths of an approximation instead of that of T. CPApc and CPAph are forward, best first search planners based on this idea. We compare them with two state-of-the-art conformant planners, KACMBP and Conformant-FF (CFF), over benchmarks in the literature, and over two new domains. One has large number of state constraints and another has a high degree of incompleteness. Our planners perform reasonably well in benchmark domains and outperform KACMBP and CFF in the first domain while still working well with the second one. Our experimental result shows that having an integral part of a conformant planner to deal with state constraints directly can significantly improve its performance extending a similar claim for classical planners in (Thiebaux. Hoffmann, & Nebel 2003).

Some Results on the Completeness of Approximation Based Reasoning We present two results that relate the completeness condi- tions for the 0-approximation for two formalisms: the action description language A and the situation calculus. The first result indicates that the completeness condition for the situa- tion calculus formalism implies the corresponding condition for the action language formalism. The second result indi- cates that an action theory in A can sometimes be simplified to an equivalent action theory whose completeness condition is weaker than the original theory for certain queries.

On the Completeness of Approximation Based Reasoning and Planning in Action Theories with Incomplete Information In this paper, we study the completeness of the 0- approximation for action theories with incomplete informa- tion. We propose a sufficient condition for which an action theory under the 0-approximation semantics is complete with respect to the possible world semantics. We then introduce the notion of decisive sets of fluents, based on which an ac- tion theory can be modified into another action theory such that the modified action theory under the 0-approximation is complete with respect to the original theory. We present a polynomial time algorithm for computing decisive sets for action theories and use it in the development of a sound and complete conformant planner. Finally, we compare our plan- ner with other state-of-the-art conformant planners.

Compiling uncertainty away: solving conformant planning problems using a classical planner (sometimes) Even under polynomial restrictions on plan length, conformant planning remains a very hard computational problem as plan verification itself can take exponential time. This heavy price cannot be avoided in general although in many cases conformant plans are verifiable efficiently by means of simple forms of disjunctive inference. This raises the question of whether it is possible to identify and use such forms of inference for developing an efficient but incomplete planner capable of solving non-trivial problems quickly. In this work, we show that this is possible by mapping conformant into classical problems that are then solved by an off-the-shelf classical planner. The formulation is sound as the classical plans obtained are all conformant, but it is incomplete as the inverse relation does not always hold. The translation accommodates 'reasoning by cases' by means of an 'split-protect-and-merge' strategy; namely, atoms L/Xi that represent conditional beliefs 'if Xi then L' are introduced in the classical encoding, that are combined by suitable actions to yield the literal L when the disjunction X1 ∨ ... ∨ Xn holds and certain invariants in the plan are verified. Empirical results over a wide variety of problems illustrate the power of the approach.

An Approximation of Action Theories of and Its Application to Conformant Planning In this paper we generalize the notion of approximation of action theories introduced in [ 13,26]. We introduce a logic programming based method for constructing approximation of action theories of AL and prove its soundness. We describe an approximation based conformant planner and compare its performance with other state-of-the-art conformant planners.

Improving Heuristics for Planning as Search in Belief Space Search in the space of beliefs has been proposed as a con- venient framework for tackling planning under uncertainty. Significant improvements have been recently achieved, espe- cially thanks to the use of symbolic model checking tech- niques such as Binary Decision Diagrams. However, the problem is extremely complex, and the heuristics available so far are unable to provide enough guidance for an informed search. In this paper we tackle the problem of defining effective heuristics for driving the search in belief space. The basic intuition is that the "degree of knowledge" associated with the belief states reached by partial plans must be explicitly taken into account when deciding the search direction. We propose a way of ranking belief states depending on their de- gree of knowledge with respect to a given set of boolean func- tions. This allows us to define a planning algorithm based on the identification and solution of suitable "knowledge sub- goals", that are used as intermediate steps during the search. The solution of knowledge subgoals is based on the identifi- cation of "knowledge acquisition conditions", i.e. subsets of the state space from where it is possible to perform knowl- edge acquisition actions. We show the effectiveness of the proposed ideas by observing substantial improvements in the conformant planning algorithms of MBP.

Adding knowledge to the action description language A We introduce Ak an extension of the action description language A (Gelfond & Lifschitz 1993) to handle actions which affect knowledge. We use sensing actions to increase an agent's knowledge of the world and non-deterministic actions to remove knowledge. We include complex plans involving conditionals and loops in our query language for hypothetical reasoning. Finally, we present a translation of descriptions in Ak to epistemic logic programs.

An Approach To Planning With Incomplete Information Classical planners presuppose complete and correct informationabout the world. This paper provides thesyntax and semantics for uwl, a representation forgoals and actions that facilitates planning with incompleteinformation about the world's state. While theexpressive power of uwl is limited compared to previouswork on logics of knowledge and belief, uwl hasthe advantage of being easily incorporated into planningalgorithms. We describe a provably correct planningalgorithm based on uwl....

Refinement Planning as a Unifying Framework for Plan Synthesis

On the compilability and expressive power of propositional planning formalisms The recent approaches of extending the GRAPHPLAN algorithm to handle more expressive planning formalisms raise the question of what the formal meaning of "expressive power" is. We formalize the intuition that expressive power is a measure of how concisely planning domains and plans can be expressed in a particular formalism by introducing the notion of "compilation schemes" between planning formalisms. Using this notion, we analyze the expressiveness of a large family of propositional planning formalisms, ranging from basic STRIPS to a formalism with conditional effects, partial state specifications, and propositional formulae in the preconditions. One of the results is that conditional effects cannot be compiled away if plan size should grow only linearly but can be compiled away if we allow for polynomial growth of the resulting plans. This result confirms that the recently proposed extensions to the GRAPHPLAN algorithm concerning conditional effects are optimal with respect to the "compilability" framework. Another result is that general propositional formulae cannot be compiled into conditional effects if the plan size should be preserved linearly. This implies that allowing general propositional formulae in preconditions and effect conditions adds another level of difficulty in generating a plan.

Exploiting generative models in discriminative classifiers Generative probability models such as hidden Markov models provide a principled way of treating missing information and dealing with variable length sequences. On the other hand, discriminative methods such as support vector machines enable us to construct flexible decision boundaries and often result in classification performance superior to that of the model based approaches. An ideal classifier should combine these two complementary approaches. In this paper, we develop a natural way of achieving this combination by deriving kernel functions for use in discriminative methods such as support vector machines fr om generative probability models. We provide a theoretical justification for this combination as well as demonstrate a substantial improvement in the classification performance in the context of DNA and protein sequence analysis.

Distributed coupled actors: A Chorus proposal for reliability

Oriented principal component analysis for large margin classifiers. Large margin classifiers (such as MLPs) are designed to assign training samples with high confidence (or margin) to one of the classes. Recent theoretical results of these systems show why the use of regularisation terms and feature extractor techniques can enhance their generalisation properties. Since the optimal subset of features selected depends on the classification problem, but also on the particular classifier with which they are used, global learning algorithms for large margin classifiers that use feature extractor techniques are desired. A direct approach is to optimise a cost function based on the margin error, which also incorporates regularisation terms for controlling capacity. These terms must penalise a classifier with the largest margin for the problem at hand. Our work shows that the inclusion of a PCA term can be employed for this purpose. Since PCA only achieves an optimal discriminatory projection for some particular distribution of data, the margin of the classifier can then be effectively controlled. We also propose a simple constrained search for the global algorithm in which the feature extractor and the classifier are trained separately. This allows a degree of flexibility for including heuristics that can enhance the search and the performance of the computed solution. Experimental results demonstrate the potential of the proposed method.

Improving Citation Polarity Classification With Product Reviews Recent work classifying citations in scientific literature has shown that it is possible to improve classification results with extensive feature engineering. While this result confirms that citation classification is feasible, there are two drawbacks to this approach: (i) it requires a large annotated corpus for supervised classification, which in the case of scientific literature is quite expensive; and (ii) feature engineering that is too specific to one area of scientific literature may not be portable to other domains, even within scientific literature. In this paper we address these two drawbacks. First, we frame citation classification as a domain adaptation task and leverage the abundant labeled data available in other domains. Then, to avoid over-engineering specific citation features for a particular scientific domain, we explore a deep learning neural network approach that has shown to generalize well across domains using unigram and bigram features. We achieve better citation classification results with this cross-domain approach than using in-domain classification.

A linear time transform for probability aware planning We present a transform that enables traditional Shortest-Feasible-Plan planners to reason about uncertain operators and produce plans which have higher probabilities of success. This transform converts a probability-aware domain description into a STRIPS-style description, where the probability of success is expressed by plan length. Using this transformed description a plan can be generated by a traditional planner. The transform is shown to be at worst linear in the size of the input, and allows the planning system to trade-off accuracy against runtime as an anytime computation.

The Complexity of Plan Existence and Evaluation in Probabilistic Domains We examine the computational complexity of testing and finding small plans inprobabilistic planning domains (both flat and succinct). We show that many problemsof interest are complete for a variety of complexity classes: PL, P, NP, co-NP, PP,NPPP, co-NPPP, and PSPACE. Of these, the probabilistic classes PP and NPPParelikely to be of special interest in the field of uncertainty in artificial intelligence andare deserving of additional study.1 IntroductionRecent work in ...

Abstraction and approximate decision-theoretic planning ion and Approximate Decision TheoreticPlanningRichard Dearden and Craig BoutilieryDepartment of Computer ScienceUniversity of British ColumbiaVancouver, British ColumbiaCANADA, V6T 1Z4email: dearden,[email protected] decision processes (MDPs) have recently been proposed asuseful conceptual models for understanding decision-theoretic planning.However, the utility of the associated computational methods remainsopen to question: most algorithms for computing optimal...

The computational complexity of probabilistic planning We examine the computational complexity of testing and finding small plans in probabilistic planning domains with both flat and propositional representations. The complexity of plan evaluation and existence varies with the plan type sought; we examine totally ordered plans, acyclic plans, and looping plans, and partially ordered plans under three natural definitions of plan value. We show that problems of interest are complete for a variety of complexity classes: PL, P, NP, co-NP, PP, NPPP, co-NPPP, and PSPACE. In the process of proving that certain planning problems are complete for NPPP, we introduce a new basic NPPP -complete problem, E-MAJSAT, which generalizes the standard Boolean satisfiability problem to computations involving probabilistic quantities; our results suggest that the development of good heuristics for E-MAJSAT could be important for the creation of efficient algorithms for a wide variety of problems.

Logic programs with classical negation

Logic programming and knowledge representation In this paper, we review recent work aimed at the application of declarative logic programming to knowledge representation in artificial intelligence. We consider extensions of the language of definite logic programs by classical (strong) negation, disjunction, and some modal operators and show how each of the added features extends the representational power of the language.

The contract net protocol: high-level communication and control in a distributed problem solver The contract net protocol has been developed to specify problem-solving communication and control for nodes in a distributed problem solver. Task distribution is affected by a negotiation process, a discussion carried on between nodes with tasks to be executed and nodes that may be able to execute those tasks.

A trace-driven analysis of the UNIX 4.2 BSD file system

Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence, IJCAI 99, Stockholm, Sweden, July 31 - August 6, 1999. 2 Volumes, 1450 pages

Fine-Grained Mobility in the Emerald System (Extended Abstract)

Representing actions in logic programs and default theories a situation calculus approach We address the problem of representing common sense knowledge about action domains in the formalisms of logic programming and default logic. We employ a methodology proposed by Gelfond and Lifschitz which involves first defining a high-level language for representing knowledge about actions, and then specifying a translation from the high-level action language into a general-purpose formalism, such as logic programming. Accordingly, we define a high-level action languageAE, and specify sound and complete translations of portions ofAEinto logic programming and default logic. The languageAEincludes propositions that represent “static causal laws” of the following kind: a fluent formula ψ can be made true by making a fluent formula true (or, more precisely, ψ is caused whenever is caused). Such propositions are more expressive than the state constraints traditionally used to represent background knowledge. Our translations ofAEdomain descriptions into logic programming and default logic are simple, in part because the noncontrapositive nature of causal laws is easily reflected in such rule-based formalisms.

iSAM: Incremental Smoothing and Mapping In this paper, we present incremental smoothing and mapping (iSAM), which is a novel approach to the simultaneous localization and mapping problem that is based on fast incremental matrix factorization. iSAM provides an efficient and exact solution by updating a QR factorization of the naturally sparse smoothing information matrix, thereby recalculating only those matrix entries that actually change. iSAM is efficient even for robot trajectories with many loops as it avoids unnecessary fill-in in the factor matrix by periodic variable reordering. Also, to enable data association in real time, we provide efficient algorithms to access the estimation uncertainties of interest based on the factored information matrix. We systematically evaluate the different components of iSAM as well as the overall algorithm using various simulated and real-world datasets for both landmark and pose-only settings.

When Multivariate Forecasting Meets Unsupervised Feature Learning - Towards a Novel Anomaly Detection Framework for Decision Support. Many organizations adopt information technologies to make intelligent decisions during operations. Time-series data plays a crucial role in supporting such decision making processes. Though current studies on time-series based decision making provide reasonably well results, the anomaly detection essence underling most of the scenarios and the plenitude of unlabeled data are largely overlooked and left unexplored. We argue that by using multivariate forecasting and unsupervised feature learning, these two important research gaps could be filled. We carried out two experiments in this study to testify our approach and the results showed that decision support performance was significantly improved. We also proposed a novel framework to integrate the two methods so that our approach may be generalized to a larger problem domain. We discussed the advantages, the limitations and the future work of our study. Both practical and theoretical contributions were also discussed in the paper. © 2012 by the AIS/ICIS Administrative Office All rights reserved.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Heterogeneous active agents, I: semantics Over the years, many different agent programming languages have been proposed. In this paper, we propose a concept called Agent Programs using which, the way an agent should act in various situations can be declaratively specified by the creator of that agent. Agent Programs may be built on top of arbitrary pieces of software code and may be used to specify what an agent is obliged to do, what an agent may do, and what an agent may not do. In this paper, we define several successively more sophisticated and epistemically satisfying declarative seman- tics for agent programs. We further show that agent programs cleanly extend well understood semantics for logic programs, and thus are clearly linked to existing res ults on logic program- ming and nonmonotonic reasoning.

Heterogeneous active agents, III: polynomially implementable agents In (17), two of the authors have introduced techniques to build agents on top of arbitrary data structures, and to "agentize" new/existing programs. They provid ed a series of succes- sively more sophisticated semantics for such agent systems, and showed that as these semantics become epistemically more desirable, a computational price may need to be paid. In this paper, we identify a class of agents that are called weakly regular—this is done by first identifying a fragment of agent programs (17) called weakly regular agent programs (WRAPs for short). It is shown that WRAPs are definable via three parameters—checking for a property called "safety", checking for a property called "conflict freedom" and checking for a "deonti c strat- ifiability" property. Algorithms for each of these are developed. A weakly regular agent is then defined in terms of these concepts , and a regular agent is one that satisfies an additional boundedness property. We then describe a polynomial algorithm that computes (under suitable assumptions) the reasonable status set semantics of regular agents—this semantics was iden- tified in (17) as being epistemically most desirable. Though this semantics is coNP-complete for arbitrary agent programs (16), it is polynomially computable via our algorithm for regular agents. Finally, we describe our implementation architecture and provide details of how we have implemented RAPs, together with experimental results.

A Deductive Database Approach to Planning in Uncertain Environments We present a formal model for reasoning about probabilistic information in STRIPS style planning. We then show that all probabilistic planning problems expressible in this model may be represented as equivalent probabilistic logic programs, yielding a sound and complete method for finding such plans.

From logic programming towards multi-agent systems In this paper we present an extension of logic programming (LP) that is suitable not only for the “rational” component of a single agent but also for the “reactive” component and that can encompass multi‐agent systems. We modify an earlier abductive proof procedure and embed it within an agent cycle. The proof procedure incorporates abduction, definitions and integrity constraints within a dynamic environment, where changes can be observed as inputs. The definitions allow rational planning behaviour and the integrity constraints allow reactive, condition‐action type behaviour. The agent cycle provides a resource‐bounded mechanism that allows the agent’s thinking to be interrupted for the agent to record and assimilate observations as input and execute actions as output, before resuming further thinking. We argue that these extensions of LP, accommodating multi‐theories embedded in a shared environment, provide the necessary multi‐agent functionality. We argue also that our work extends Shoham’s Agent0 and the BDI architecture.

A Policy Description Language A policy describes principles or strategies for a plan of action designed to achieve a particular set of goals. We define a policy as a function that maps a series of events into a set of actions. In this paper we introduce PDL, a simple but expressive language to specify policies. The design of the language has been strongly influenced by the action languages of Geffner and Bonet (Geffner

Perspectives on artificial intelligence planning Planning is a key area in Artificial Intelligence. In its general form, planning is concerned with the automatic synthesis of action strategies (plans) from a description of actions, sensors, and goals. Planning thus contrasts with two other approaches to intelligent behavior: the programming approach, where action strategies are defined by hand, and the learning approach, where action strategies are inferred from experience. Different assumptions about the nature of actions, sensors, and costs lead to various forms of planning: planning with complete information and deterministic actions (classical planning), planning with non-deterministic actions and sensing, planning with temporal and concurrent actions, etc. Most work so far has been devoted to classical planning, where significant changes have taken place in the last few years. On the methodological side, the area has become more empirical, on the technical side, approaches based on heuristic or constrained-based search have become common.In this paper, I try to provide a coherent picture of Planning in AI, making emphasis on the mathematical models that underlie various forms of planning and the ideas that have been found most useful computationally.

Towards a general theory of action and time A formalism for reasoning about actions is proposed that is based on a temporal logic. It allows a much wider range of actions to be described than with previous approaches such as the situation calculus. This formalism is then used to characterize the different types of events, processes, actions, and properties that can be described in simple English sentences. In addressing this problem, we consider actions that involve non-activity as well as actions that can only be defined in terms of the beliefs and intentions of the actors. Finally, a framework for planning in a dynamic world with external events and multiple agents is suggested.

Logic programs with exceptions We extend logic programming to deal with default reasoning by allowing the explicit representation of exceptions in addition to general rules. To formalise this extension, we modify the answer set semantics of Gelfond and Lifschitz, which allows both classical negation and negation as failure. We also propose a transformation which eliminates exceptions by using negation by failure. The transformed program can be implemented by standard logic programming methods, such as SLDNF. The explicit representation of rules and exceptions has the virtue of greater naturalness of expression. The transformed program, however, is easier to implement.

Heuristics based on unit propagation for satisfiability problems The paper studies new unit propagation based heuristics for Davis-Putnam-Loveland (DPL) procedure. These are the novel combinations of unit propagation and the usual "Maximum Occurrences in clauses of Minimum Size" heuristics. Based on the experimental evaluations of different alternatives a new simple unit propagation based heuristic is put forward. This compares favorably with the heuristics employed in the current state-of-the-art DPL implementations (C-SAT, Tableau, POSIT).

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Exokernel: an operating system architecture for application-level resource management Traditional operating systems limit the performance, flexibility, and functionality of applications by fixing the interface and implementation of operating system abstractions such as interprocess communication and virtual memory. The exokernel operating system architecture addresses this problem by providing application-level management of physical resources. In the exokernel architecture, a small kernel securely exports all hardware resources through a lowlevel interface to untrusted library operating systems. Library operating systems use this interface to implement system objects and policies. This separation of resource protection from management allows application-specific customization of traditional operating system abstractions by extending, specializing, or even replacing libraries. We have implemented a prototype exokernel operating system. Measurements show that most primitive kernel operations (such as exception handling and protected control transfer) are ten to 100 times faster than in Ultrix, a mature monolithic UNIX operating system. In addition, we demonstrate that an exokemel allows applications to control machine resources in ways not possible in traditional operating systems. For instance, virtual memory and interprocess communication abstractions are implemented entirely within an application-level library. Measurements show that application-level virtual memory and interprocess communication primitives are five to 40 times faster than Ultrix’s kernel primitives. Compared to state-of-the-art implementations from the literature, the prototype exokemel system is at least five times faster on operations such as exception dispatching and interprocess communication.

For the sake of simplicity: unsupervised extraction of lexical simplifications from Wikipedia We report on work in progress on extracting lexical simplifications (e.g., "collaborate" → "work together"), focusing on utilizing edit histories in Simple English Wikipedia for this task. We consider two main approaches: (1) deriving simplification probabilities via an edit model that accounts for a mixture of different operations, and (2) using metadata to focus on edits that are more likely to be simplification operations. We find our methods to outperform a reasonable baseline and yield many high-quality lexical simplifications not included in an independently-created manually prepared list.

Planning as satisfiability: Heuristics Reduction to SAT is a very successful approach to solving hard combinatorial problems in Artificial Intelligence and computer science in general. Most commonly, problem instances reduced to SAT are solved with a general-purpose SAT solver. Although there is the obvious possibility of improving the SAT solving process with application-specific heuristics, this has rarely been done successfully. In this work we propose a planning-specific variable selection strategy for SAT solving. The strategy is based on generic principles about properties of plans, and its performance with standard planning benchmarks often substantially improves on generic variable selection heuristics, such as VSIDS, and often lifts it to the same level with other search methods such as explicit state-space search with heuristic search algorithms.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Domain adaptation problems: a DASVM classification technique and a circular validation strategy. This paper addresses pattern classification in the framework of domain adaptation by considering methods that solve problems in which training data are assumed to be available only for a source domain different (even if related) from the target domain of (unlabeled) test data. Two main novel contributions are proposed: 1) a domain adaptation support vector machine (DASVM) technique which extends the formulation of support vector machines (SVMs) to the domain adaptation framework and 2) a circular indirect accuracy assessment strategy for validating the learning of domain adaptation classifiers when no true labels for the target--domain instances are available. Experimental results, obtained on a series of two-dimensional toy problems and on two real data sets related to brain computer interface and remote sensing applications, confirmed the effectiveness and the reliability of both the DASVM technique and the proposed circular validation strategy.

Distant Domain Transfer Learning. In this paper, we study a novel transfer learning problem termed Distant Domain Transfer Learning (DDTL). Different from existing transfer learning problems which assume that there is a close relation between the source domain and the target domain, in the DDTL problem, the target domain can be totally different from the source domain. For example, the source domain classifies face images but the target domain distinguishes plane images. Inspired by the cognitive process of human where two seemingly unrelated concepts can be connected by learning intermediate concepts gradually, we propose a Selective Learning Algorithm (SLA) to solve the DDTL problem with supervised autoencoder or supervised convolutional autoencoder as a base model for handling different types of inputs. Intuitively, the SLA algorithm selects usefully unlabeled data gradually from intermediate domains as a bridge to break the large distribution gap for transferring knowledge between two distant domains. Empirical studies on image classification problems demonstrate the effectiveness of the proposed algorithm, and on some tasks the improvement in terms of the classification accuracy is up to 17% over "non-transfer" methods.

Efficient Learning of Domain-invariant Image Representations

Using Different Cost Functions to Train Stacked Auto-Encoders Deep neural networks comprise several hidden layers of units, which can be pre-trained one at a time via an unsupervised greedy approach. A whole network can then be trained (fine-tuned) in a supervised fashion. One possible pre-training strategy is to regard each hidden layer in the network as the input layer of an auto-encoder. Since auto-encoders aim to reconstruct their own input, their training must be based on some cost function capable of measuring reconstruction performance. Similarly, the supervised fine-tuning of a deep network needs to be based on some cost function that reflects prediction performance. In this work we compare different combinations of cost functions in terms of their impact on layer-wise reconstruction performance and on supervised classification performance of deep networks. We employed two classic functions, namely the cross-entropy (CE) cost and the sum of squared errors (SSE), as well as the exponential (EXP) cost, inspired by the error entropy concept. Our results were based on a number of artificial and real-world data sets.

A SVM-based model-transferring method for heterogeneous domain adaptation. In many real classification scenarios the distribution of test (target) domain is different from the training (source) domain. The distribution shift between the source and target domains may cause the source classifier not to gain the expected accuracy on the target data. Domain adaptation has been introduced to solve the accuracy-dropping problem caused by distribution shift phenomenon between domains. In this paper, we study model-transferring methods as a practical branch of adaptation methods, which adapt the source classifier to new domains without using the source samples. We introduce a new SVM-based model-transferring method, in which a max-margin classifier is trained on labeled target samples and is adapted using the offset of the source classifier. We call it Heterogeneous Max-Margin Classifier Adaptation Method, abbreviated as HMCA. The main strength of HMCA is its applicability for heterogeneous domains where the source and target domains may have different feature types. This property is important because the previously proposed model-transferring methods do not provide any solution for heterogeneous problems. We also introduce a new similarity metric that reliably measures adaptability between two domains according to HMCA structure. In the situation that we have access to several source classifiers, the metric can be used to select the most appropriate one for adaptation. We test HMCA on two different computer vision problems (pedestrian detection and image classification). The experimental results show the advantage in accuracy rate for our approach in comparison to several baselines. We propose a new SVM-based model-transferring method for adaptation.Our method applies adaptation in the one-dimensional discrimination space.The proposed method can handle heterogeneous domains.Based on proposed model-transferring method, we design a new metric for measuring the adaptability between two domains.

Domain Adaptive Classification We propose an unsupervised domain adaptation method that exploits intrinsic compact structures of categories across different domains using binary attributes. Our method directly optimizes for classification in the target domain. The key insight is finding attributes that are discriminative across categories and predictable across domains. We achieve a performance that significantly exceeds the state-of-the-art results on standard benchmarks. In fact, in many cases, our method reaches the same-domain performance, the upper bound, in unsupervised domain adaptation scenarios.

Representation learning: a review and new perspectives. The success of machine learning algorithms generally depends on data representation, and we hypothesize that this is because different representations can entangle and hide more or less the different explanatory factors of variation behind the data. Although specific domain knowledge can be used to help design representations, learning with generic priors can also be used, and the quest for AI is motivating the design of more powerful representation-learning algorithms implementing such priors. This paper reviews recent work in the area of unsupervised feature learning and deep learning, covering advances in probabilistic models, autoencoders, manifold learning, and deep networks. This motivates longer term unanswered questions about the appropriate objectives for learning good representations, for computing representations (i.e., inference), and the geometrical connections between representation learning, density estimation, and manifold learning.

Exploiting generative models in discriminative classifiers Generative probability models such as hidden Markov models provide a principled way of treating missing information and dealing with variable length sequences. On the other hand, discriminative methods such as support vector machines enable us to construct flexible decision boundaries and often result in classification performance superior to that of the model based approaches. An ideal classifier should combine these two complementary approaches. In this paper, we develop a natural way of achieving this combination by deriving kernel functions for use in discriminative methods such as support vector machines fr om generative probability models. We provide a theoretical justification for this combination as well as demonstrate a substantial improvement in the classification performance in the context of DNA and protein sequence analysis.

Differentiable Sparse Coding Prior work has shown that features which appear to be biologically plausible as well as empirically useful can be found by sparse coding with a prior such as a laplacian (L1) that promotes sparsity. We show how smoother priors can pre- serve the benefits of these sparse priors while adding stability to the Maximum A-Posteriori (MAP) estimate that makes it more useful for prediction problems. Additionally, we show how to calculate the derivative of the MAP estimate effi- ciently with implicit differentiation. One prior that can be differentiated this way is KL-regularization. We demonstrate its effectiveness on a wide variety of appli- cations, and find that online optimization of the parameters of the KL-regularized model can significantly improve prediction performance.

Statistical Parametric Speech Synthesis Using Deep Neural Networks Conventional approaches to statistical parametric speech synthesis typically use decision tree-clustered context-dependent hidden Markov models (HMMs) to represent probability densities of speech parameters given texts. Speech parameters are generated from the probability densities to maximize their output probabilities, then a speech waveform is reconstructed from the generated parameters. This approach is reasonably effective but has a couple of limitations, e.g. decision trees are inefficient to model complex context dependencies. This paper examines an alternative scheme that is based on a deep neural network (DNN). The relationship between input texts and their acoustic realizations is modeled by a DNN. The use of the DNN can address some limitations of the conventional approach. Experimental results show that the DNN-based systems outperformed the HMM-based systems with similar numbers of parameters.

Probabilistic quantifiers and games We consider inclusion relations among a multitude of classical complexity classes and classes with probabilistic components. A key tool is a method for characterizing such classes in terms of the ordinary quantifiers ∃ and ∠ together with a quantifier ∃ + , which means roughly “for most,” applied to polynomial-time predicates. This approach yields a uniform treatment which leads to easier proofs for class-inclusion and hierarchy-collapse results. Furthermore, the method captures some recently introduced game classes and game hierarchies. This survey also includes a charting of class-inclusion and oracle-based separation results.

ADL and the State-Transition Model of Action

Experimenting with power default reasoning In this paper we explore the computational aspects of Propositional Power Default Reasoning (PDR), a form of non-monotonic reasoning in which the underlying logic is Kleene's 3-valued propositional logic. PDR leads to a concise meaning of the problem of skeptical entailment which has better complexity characteristics than the usual formalisms (co-NP(3)-Complete instead of 驴2P-Complete). We take advantage of this in an implementation called powdef to encode and solve hard graph problems and explore randomly generated instances of skeptical entailment.

"The sum of all human knowledge": A systematic review of scholarly research on the content of Wikipedia AbstractWikipedia may be the best-developed attempt thus far to gather all human knowledge in one place. Its accomplishments in this regard have made it a point of inquiry for researchers from different fields of knowledge. A decade of research has thrown light on many aspects of the Wikipedia community, its processes, and its content. However, due to the variety of fields inquiring about Wikipedia and the limited synthesis of the extensive research, there is little consensus on many aspects of Wikipedia's content as an encyclopedic collection of human knowledge. This study addresses the issue by systematically reviewing 110 peer-reviewed publications on Wikipedia content, summarizing the current findings, and highlighting the major research trends. Two major streams of research are identified: the quality of Wikipedia content including comprehensiveness, currency, readability, and reliability and the size of Wikipedia. Moreover, we present the key research trends in terms of the domains of inquiry, research design, data source, and data gathering methods. This review synthesizes scholarly understanding of Wikipedia content and paves the way for future studies.

CUDA-MEME: Accelerating motif discovery in biological sequences using CUDA-enabled graphics processing units Motif discovery in biological sequences is of prime importance and a major challenge in computational biology. Consequently, numerous motif discovery tools have been developed to date. However, the rapid growth of both genomic sequence and gene transcription data, establishes the need for the development of scalable motif discovery tools. An approach to improve the runtime of motif discovery by an order-of-magnitude without losing sensitivity is to employ emerging many-core architectures such as CUDA-enabled GPUs. In this paper, we present a highly parallel formulation and implementation of the MEME motif discovery algorithm using the CUDA programming model. To achieve high efficiency, we introduce two parallelization approaches: sequence-level and substring-level parallelization. Furthermore, a hybrid computing framework is described to take advantage of both CPU and GPU compute resources. Our performance evaluation on a GeForce GTX 280 GPU, results in average runtime speedups of 21.4 (19.3) for the starting point search and 20.5 (16.4) for the overall runtime using the OOPS (ZOOPS) motif search model. The runtime speedups of CUDA-MEME on a single GPU are also comparable to those of ParaMEME running on 16 CPU cores of a high-performance workstation cluster. In addition to the fast speed, CUDA-MEME has the capability of finding motif instances consistent with the sequential MEME.

Smith-Waterman implementation on a FSB-FPGA module using the Intel Accelerator Abstraction Layer The Smith-Waterman algorithm is employed in the field of Bioinformatics to find optimal local alignments of two DNA or protein sequences. It is a classic example of a dynamic programming algorithm. Because it is highly parallel both spatially and temporally and because the fundamental data structure is compact, Smith-Waterman lends itself very well to operation on an FPGA. Here we demonstrate an implementation of this important algorithm in a novel FSB module using the Intel Accelerator Abstraction Layer (AAL), a newly released software middleware layer. We have modified SSEARCH35, an industry standard open-source implementation of the Smith-Waterman algorithm, to transparently introduce a hardware accelerated option to users. We demonstrate performance of nine billion cell updates per second and discuss further opportunities for performance improvement.

A Real-World Attack Breaking A5/1 within Hours In this paper we present a real-world hardware-assisted attack on the well-known A5/1 stream cipher which is (still) used to secure GSM communication in most countries all over the world. During the last ten years A5/1 has been intensively analyzed [1,2,3,4,5,6,7]. However, most of the proposed attacks are just of theoretical interest since they lack from practicability -- due to strong preconditions, high computational demands and/or huge storage requirements -- or have never been fully implemented.In contrast to these attacks, our attack which is based on the work by Keller and Seitz [8] is running on an existing special-purpose hardware device, called COPACOBANA [9]. With the knowledge of only 64 bits of keystream the machine is able to reveal the corresponding internal 64-bit state of the cipher in about 6 hours on average. We provide a detailed description of our attack architecture as well as implementation results.

Massively Parallelized DNA Motif Search on the Reconfigurable Hardware Platform COPACOBANA An enhanced version of an existing motif search algorithm BMA is presented. Motif searching is a computationally expensive task which is frequently performed in DNA sequence analysis. The algorithm has been tailored to fit on the COPACOBANA architecture, which is a massively parallel machine consisting of 120 FPGA chips. The performance gained exceeds that of a standard PC by a factor of over 1,650 and speeds up the time intensive search for motifs in DNA sequences. In terms of energy consumption COPACOBANA needs 1/400 of the energy of a PC implementation.

Efficient parallel algorithm for multiple sequence alignments with regular expression constraints on graphics processing units Multiple sequence alignments with constraints has become an important problem in computational biology. The concept of constrained sequence alignment is proposed to incorporate the biologist's domain knowledge into sequence alignments such that the user-specified residues/segments are aligned together in the alignment results. Over the past decade, a series of constrained multiple sequence alignment tools were proposed in the literature. RE-MuSiC is the newest tool with the regular expression constraints and useful for a wide range of biological applications. However, the computation time of REMuSiC is large for a large amount of sequences or long sequences and this problem limits the application usage. Therefore, in this paper, a tool, GPU-REMuSiC v1.0, is proposed to reduce the computation time of RE-MuSiC by using the graphics processing units with CUDA. GPU-REMuSiC v1.0 can achieve 29× speedups for overall computation time by the experimental results.

Accelerating BLASTP on the Cell Broadband Engine The enormous growth of biological sequence databases has caused bioinformatics to be rapidly moving towards a data-intensive, computational science. As a result, the computational power needed by bioinformatics applications is growing rapidly as well. The recent emergence of low cost parallel accelerator technologies has made it possible to reduce execution times of many bioinformatics applications. In this paper, we demonstrate how the PlayStation®3, powered by the Cell Broadband Engine, can be used as an efficient computational platform to accelerate the popular BLASTP algorithm.

Use of receiver operating characteristic (ROC) analysis to evaluate sequence matching In this paper, we borrow the idea of the receiver operating characteristic (ROC) from clinical medicine and demonstrate its application to sequence comparison. The ROC includes elements of both sensitivity and specificity, and is a quantitative measure of the usefulness of a diagnostic. The ROC is used in this work to investigate the effects of scoring table and gap penalties on database searches. Studies on three families of proteins, 4Fe-4S ferredoxins, lysR bacterial regulatory proteins, and bacterial RNA polymerase σ-factors lead to the following conclusions: sequence families are quite idiosyncratic, but the best PAM distance for database searches using the Smith-Waterman method is somewhat larger than predicted by theoretical methods, about 200 PAM. The length independent gap penalty (gap initation penalty) is quite important, but shows a broad peak at values of about 20–24. The length dependent gap penalty (gap extension penalty) is almost irrelevant suggesting that successful database searches rely only to a limited degree on gapped alignments. Taken together, these observations lead to the conclusion that the optimal conditions for alignments and database searches are not, and should not be expected to be, the same.

Acceleration of ungapped extension in Mercury BLAST The amount of biosequence data being produced each year is growing exponentially. Extracting useful information from this massive amount of data efficiently is becoming an increasingly difficult task. There are many available software tools that molecular biologists use for comparing genomic data. This paper focuses on accelerating the most widely used such tool, BLAST. Mercury BLAST takes a streaming approach to the BLAST computation by offloading the performance-critical sections to specialized hardware. This hardware is then used in combination with the processor of the host system to deliver BLAST results in a fraction of the time of the general-purpose processor alone. This paper presents the design of the ungapped extension stage of Mercury BLAST. The architecture of the ungapped extension stage is described along with the context of this stage within the Mercury BLAST system. The design is compact and runs at 100MHz on available FPGAs, making it an effective and powerful component for accelerating biosequence comparisons. The performance of this stage is 25x that of the standard software distribution, yielding close to 50x performance improvement on the complete BLAST application. The sensitivity is essentially equivalent to that of the standard distribution.

Computing Circumscription Revisited: A Reduction Algorithm In recent years, a great deal of attention has been devoted to logics of common-sense reasoning. Among the candidates proposed, circumscription has been perceived as an elegant mathematical technique for modeling nonmonotonic reasoning, but difficult to apply in practice. The major reason for this is the second-order nature of circumscription axioms and the difficulty in finding proper substitutions of predicate expressions for predicate variables. One solution to this problem is to compile, where possible, second-order formulas into equivalent first-order formulas. Although some progress has been made using this approach, the results are not as strong as one might desire and they are isolated in nature. In this article, we provide a general method that can be used in an algorithmic manner to reduce certain circumscription axioms to first-order formulas. The algorithm takes as input an arbitrary second-order formula and either returns as output an equivalent first-order formula, or terminates with failure. The class of second-order formulas, and analogously the class of circumscriptive theories that can be reduced, provably subsumes those covered by existing results. We demonstrate the generality of the algorithm using circumscriptive theories with mixed quantifiers (some involving Skolemization), variable constants, nonseparated formulas, and formulas with n-ary predicate variables. In addition, we analyze the strength of the algorithm, compare it with existing approaches, and provide formal subsumption results.

On Computing Boolean Connectives of Characteristic Functions This paper is a study of the existence of polynomial time Boolean connective functions for languages. A language L has an AND function if there is a polynomial time f such that f(x, y) ∈ L ⇐⇒ x ∈ L and y ∈ L. L has an OR function if there is a polynomial time g such that g(x, y) ∈ L ⇐⇒ x ∈ L or y ∈ L. While all NP complete sets have these functions, Graph Isomorphism, which is probably not complete, is also shown to have both AND and OR functions. The results in this paper characterize the complete sets for the classes DP and PSAT(O(log n)) in terms of AND and OR, and relate these functions to the structure of the Boolean hierarchy and the query hierarchies. Also, this paper shows that the complete sets for the levels of the Boolean hierarchy above the second level cannot have AND or OR unless the polynomial hierarchy collapses. Finally, most of the structural properties of the Boolean hierarchy and query hierarchies are shown to depend only on the existence of AND and OR functions for the NP complete sets.

Beyond Objects: Objects Object-orientation offers more than just objects, classes and inheritance as means to structure applications. It is an approach to application development in which software systems can be constructed by composing and refining pre-designed, plug- compatible software components. But for this approach to be successfully applied, programming languages must provide better support for component specification and software composition, the software development life-cycle must separate the issues of generic component design and reuse from that of constructing applications to meet specific requirements, and, more generally, the way we develop, manage, exchange and market software must adapt to better support large-scale reuse for software communities. In this paper we shall explore these themes and we will highlight a number of key research di- rections and open problems to be explored as steps towards improving the effectiveness of object technology.

On the Equivalence of XML Patterns Patterns for matching parts of XML documents are used in a number of areas of XML document management: in links between documents, in templates for document transformation, and in queries for document retrieval. The W3C has defined XSLT patterns as a common sub-language for all these applications. We study the equivalence problem for XSLT patterns by defining a logic-based data model for XML and a semantics for XSLT patterns in terms of Datalog programs. Although uniform equivalence of Datalog programs is not sufficient to capture the equivalence of programs derived from XSLT patterns, we nevertheless show that equivalence can be decided by a variant of the chase process using embedded tuple-generating dependencies. One advantage of this approach is that the method can easily be extended to determine equivalence when documents are known to satisfy constraints imposed by document type definitions.

Hot Block Clustering for Disk Arrays with Dynamic Striping

Learning A Lexical Simplifier Using Wikipedia In this paper we introduce a new lexical simplification approach. We extract over 30K candidate lexical simplifications by identifying aligned words in a sentence-aligned corpus of English Wikipedia with Simple English Wikipedia. To apply these rules, we learn a feature-based ranker using SVMnk trained on a set of labeled simplifications collected using Amazon's Mechanical Turk. Using human simplifications for evaluation, we achieve a precision of 76% with changes in 86% of the examples.

CURL: Image Classification using co-training and Unsupervised Representation Learning. Abstract In this paper we propose a strategy for semi-supervised image classification that leverages unsupervised representation learning and co-training. The strategy, that is called CURL from co-training and unsupervised representation learning, iteratively builds two classifiers on two different views of the data. The two views correspond to different representations learned from both labeled and unlabeled data and differ in the fusion scheme used to combine the image features. To assess the performance of our proposal, we conducted several experiments on widely used data sets for scene and object recognition. We considered three scenarios (inductive, transductive and self-taught learning) that differ in the strategy followed to exploit the unlabeled data. As image features we considered a combination of GIST, PHOG, and LBP as well as features extracted from a Convolutional Neural Network. Moreover, two embodiments of CURL are investigated: one using Ensemble Projection as unsupervised representation learning coupled with Logistic Regression, and one based on LapSVM. The results show that CURL clearly outperforms other supervised and semi-supervised learning methods in the state of the art.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Development of Distributed Applications with Separation of Concerns We propose an object-oriented approach to the development of distributed applications emphasizing separation of concerns. Our approach combines the needs of transparency, encapsulation of distribution issues, and support for non-traditional models, where cooperation and sharing are desired. The development process is constructive, thus allowing partial verification of results. We recognize seven concerns: fragmentation, replication, naming, concurrency, failure, configuration, and communication. Each concern is perceived in three levels of abstraction: model, policy and mechanism. We also propose an integration process centered on development stages. Both concerns and stage perspectives are part of an integrated and flexible development process.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

The student-t mixture as a natural image patch prior with application to image compression Recent results have shown that Gaussian mixture models (GMMs) are remarkably good at density modeling of natural image patches, especially given their simplicity. In terms of log likelihood on real-valued data they are comparable with the best performing techniques published, easily outperforming more advanced ones, such as deep belief networks. They can be applied to various image processing tasks, such as image denoising, deblurring and inpainting, where they improve on other generic prior methods, such as sparse coding and field of experts. Based on this we propose the use of another, even richer mixture model based image prior: the Student-t mixture model (STM). We demonstrate that it convincingly surpasses GMMs in terms of log likelihood, achieving performance competitive with the state of the art in image patch modeling. We apply both the GMM and STM to the task of lossy and lossless image compression, and propose efficient coding schemes that can easily be extended to other unsupervised machine learning models. Finally, we show that the suggested techniques outperform JPEG, with results comparable to or better than JPEG 2000.

A Deep and Tractable Density Estimator. The Neural Autoregressive Distribution Estimator (NADE) and its real-valued version RNADE are competitive density models of multidimensional data across a variety of domains. These models use a fixed, arbitrary ordering of the data dimensions. One can easily condition on variables at the beginning of the ordering, and marginalize out variables at the end of the ordering, however other inference tasks require approximate inference. In this work we introduce an efficient procedure to simultaneously train a NADE model for each possible ordering of the variables, by sharing parameters across all these models. We can thus use the most convenient model for each inference task at hand, and ensembles of such models with different orderings are immediately available. Moreover, unlike the original NADE, our training procedure scales to deep models. Empirically, ensembles of Deep NADE models obtain state of the art density estimation performance.

How to Construct Deep Recurrent Neural Networks. In this paper, we explore different ways to extend a recurrent neural network (RNN) to a \textit{deep} RNN. We start by arguing that the concept of depth in an RNN is not as clear as it is in feedforward neural networks. By carefully analyzing and understanding the architecture of an RNN, however, we find three points of an RNN which may be made deeper; (1) input-to-hidden function, (2) hidden-to-hidden transition and (3) hidden-to-output function. Based on this observation, we propose two novel architectures of a deep RNN which are orthogonal to an earlier attempt of stacking multiple recurrent layers to build a deep RNN (Schmidhuber, 1992; El Hihi and Bengio, 1996). We provide an alternative interpretation of these deep RNNs using a novel framework based on neural operators. The proposed deep RNNs are empirically evaluated on the tasks of polyphonic music prediction and language modeling. The experimental result supports our claim that the proposed deep RNNs benefit from the depth and outperform the conventional, shallow RNNs.

Learning Deep Architectures for AI Theoretical results suggest that in order to learn the kind of com-plicated functions that can represent high-level abstractions (e.g., invision, language, and other AI-level tasks), one may needdeep architec-tures. Deep architectures are composed of multiple levels of non-linearoperations, such as in neural nets with many hidden layers or in com-plicated propositional formulae re-using many sub-formulae. Searchingthe parameter space of deep architectures is a difficult task, but learningalgorithms such as those for Deep Belief Networks have recently beenproposed to tackle this problem with notable success, beating the state-of-the-art in certain areas. This monograph discusses the motivationsand principles regarding learning algorithms for deep architectures, inparticular those exploiting as building blocks unsupervised learning ofsingle-layer models such as Restricted Boltzmann Machines, used toconstruct deeper models such as Deep Belief Networks.

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Feasibility of a serverless distributed file system deployed on an existing set of desktop PCs We consider an architecture for a serverless distributed file system that does not assume mutual trust among the client computers. The system provides security, availability, and reliability by distributing multiple encrypted replicas of each file among the client machines. To assess the feasibility of deploying this system on an existing desktop infrastructure, we measure and analyze a large set of client machines in a commercial environment. In particular, we measure and report results on disk usage and content; file activity; and machine uptimes, lifetimes, and loads. We conclude that the measured desktop infrastructure would passably support our proposed system, providing availability on the order of one unfilled file request per user per thousand days.

Partitioning and Mapping Algorithms into Fixed Size Systolic Arrays A technique for partitioning and mapping algorithms into VLSI systolic arrays is presented in this paper. Algorithm partitioning is essential when the size of a computational problem is larger than the size of the VLSI array intended for that problem. Computational models are introduced for systolic arrays and iterative algorithms. First, we discuss the mapping of algorithms into arbitrarily large size VLSI arrays. This mapping is based on the idea of algorithm transformations. Then, we present an approach to algorithm partitioning which is also based on algorithm transformations. Our approach to the partitioning problem is to divide the algorithm index set into bands and to map these bands into the processor space. The partitioning and mapping technique developed throughout the paper is summarized as a six step procedure. A computer program implementing this procedure was developed and some results obtained with this program are presented.

Indexing By Latent Semantic Analysis

Disk Shadowing Disk shadowing is a technique for maintaining a set of two or more identical disk images on separate disk devices. Its primary purpose is to enhance reliability and availability of secondary storage by providing multiple paths to redundant data. However, shadowing can also boost I/O performance. In this paper, we contend that intelligent device scheduling of shadowed discs increases the I/O rate by allowing parallel reads and by substantially reducing the average seek time for random reads. In particular, we develop and analytic model which shows that the seek time for a random read in a shadow set is a monotonic decreasing function of the number of disks.

Downward Separation Fails Catastrophically for Limited Nondeterminism Classes The $\beta$ hierarchy consists of classes $\beta_k={\rm NP}[logkn]\subseteq {\rm NP}$. Unlike collapses in the polynomial hierarchy and the Boolean hierarchy, collapses in the $\beta$ hierarchy do not seem to translate up, nor does closure under complement seem to cause the hierarchy to collapse. For any consistent set of collapses and separations of levels of the hierarchy that respects ${\rm P} = \beta_1\subseteq \beta_2\subseteq \cdots \subseteq {\rm NP}$, we can construct an oracle relative to which those collapses and separations hold; at the same time we can make distinct levels of the hierarchy closed under computation or not, as we wish. To give two relatively tame examples: for any $k \geq 1$, we construct an oracle relative to which \[ {\rm P} = \beta_{k} \neq \beta_{k+1} \neq \beta_{k+2} \neq \cdots \] and another oracle relative to which \[ {\rm P} = \beta_{k} \neq \beta_{k+1} = {\rm PSPACE}. \] We also construct an oracle relative to which $\beta_{2k} = \beta_{2k+1} \neq \beta_{2k+2}$ for all k.

Normal forms for answer sets programming Normal forms for logic programs under stable/answer set semantics are introduced. We argue that these forms can simplify the study of program properties, mainly consistency. The first normal form, called the kernel of the program, is useful for studying existence and number of answer sets. A kernel program is composed of the atoms which are undefined in the Well-founded semantics, which are those that directly affect the existence of answer sets. The body of rules is composed of negative literals only. Thus, the kernel form tends to be significantly more compact than other formulations. Also, it is possible to check consistency of kernel programs in terms of colorings of the Extended Dependency Graph program representation which we previously developed. The second normal form is called 3-kernel. A 3-kernel program is composed of the atoms which are undefined in the Well-founded semantics. Rules in 3-kernel programs have at most two conditions, and each rule either belongs to a cycle, or defines a connection between cycles. 3-kernel programs may have positive conditions. The 3-kernel normal form is very useful for the static analysis of program consistency, i.e. the syntactic characterization of existence of answer sets. This result can be obtained thanks to a novel graph-like representation of programs, called Cycle Graph which presented in the companion article Costantini (2004b).

ARIMA time series modeling and forecasting for adaptive I/O prefetching Bursty application I/O patterns, together with transfer limited storage devices, combine to create a major I/O bottleneck on parallel systems. This paper explores the use of time series models to forecast application I/O request times, then prefetching I/O requests during computation intervals to hide I/O latency. Experimental results with I/O intensive scientific codes show performance improvements compared to standard UNIX prefetching strategies.

Scheduling parallel I/O operations The I/O bottleneck in parallel computer systems has recently begun receiving increasing interest. Most attention has focused on improving the performance of I/O devices using fairly low-level parallelism in techniques such as disk striping and interleaving. Widely applicable solutions, however, will require an integrated approach which addresses the problem at multiple system levels, including applications, systems software, and architecture. We propose that within the context of such an integrated approach, scheduling parallel I/O operations will become increasingly attractive and can potentially provide substantial performance benefits.We describe a simple I/O scheduling problem and present approximate algorithms for its solution. The costs of using these algorithms in terms of execution time, and the benefits in terms of reduced time to complete a batch of I/O operations, are compared with the situations in which no scheduling is used, and in which an optimal scheduling algorithm is used. The comparison is performed both theoretically and experimentally. We have found that, in exchange for a small execution time overhead, the approximate scheduling algorithms can provide substantial improvements in I/O completion times.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Logical Vision: Meta-Interpretive Learning for Simple Geometrical Concepts.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Toward a Sequential Approach to Pipelined Image Recognition This paper introduces a sequentially motivated approach to processing streams of images from datasets with low memory demands. We utilize fuzzy clustering as an incremental dictionary learning scheme and explain how the corresponding membership functions can be subsequently used in encoding features for image patches. We focus on replicating the codebook learning and classification stages from an established visual learning pipeline that has recently shown efficacy on the CIFAR-10 small image dataset. Experiments show that performance near batch oriented learning is achievable by combining naturally online learning mechanisms driven largely by stochastic gradient descent with strictly patch-wise operations. We further detail how back propagation can be used with a neural network classifier to modify parameters within the pipeline.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Exact solutions to the nonlinear dynamics of learning in deep linear neural networks. Despite the widespread practical success of deep learning methods, our theoretical understanding of the dynamics of learning in deep neural networks remains quite sparse. We attempt to bridge the gap between the theory and practice of deep learning by systematically analyzing learning dynamics for the restricted case of deep linear neural networks. Despite the linearity of their input-output map, such networks have nonlinear gradient descent dynamics on weights that change with the addition of each new hidden layer. We show that deep linear networks exhibit nonlinear learning phenomena similar to those seen in simulations of nonlinear networks, including long plateaus followed by rapid transitions to lower error solutions, and faster convergence from greedy unsupervised pretraining initial conditions than from random initial conditions. We provide an analytical description of these phenomena by finding new exact solutions to the nonlinear dynamics of deep learning. Our theoretical analysis also reveals the surprising finding that as the depth of a network approaches infinity, learning speed can nevertheless remain finite: for a special class of initial conditions on the weights, very deep networks incur only a finite, depth independent, delay in learning speed relative to shallow networks. We show that, under certain conditions on the training data, unsupervised pretraining can find this special class of initial conditions, while scaled random Gaussian initializations cannot. We further exhibit a new class of random orthogonal initial conditions on weights that, like unsupervised pre-training, enjoys depth independent learning times. We further show that these initial conditions also lead to faithful propagation of gradients even in deep nonlinear networks, as long as they operate in a special regime known as the edge of chaos.

Provable ICA with Unknown Gaussian Noise, and Implications for Gaussian Mixtures and Autoencoders. We present a new algorithm for independent component analysis which has provable performance guarantees. In particular, suppose we are given samples of the form $$y = Ax + \\eta $$y=Ax+¿ where $$A$$A is an unknown but non-singular $$n \\times n$$n×n matrix, $$x$$x is a random variable whose coordinates are independent and have a fourth order moment strictly less than that of a standard Gaussian random variable and $$\\eta $$¿ is an $$n$$n-dimensional Gaussian random variable with unknown covariance $$\\varSigma $$Σ: We give an algorithm that provably recovers $$A$$A and $$\\varSigma $$Σ up to an additive $$\\epsilon $$∈ and whose running time and sample complexity are polynomial in $$n$$n and $$1 / \\epsilon $$1/∈. To accomplish this, we introduce a novel \"quasi-whitening\" step that may be useful in other applications where there is additive Gaussian noise whose covariance is unknown. We also give a general framework for finding all local optima of a function (given an oracle for approximately finding just one) and this is a crucial step in our algorithm, one that has been overlooked in previous attempts, and allows us to control the accumulation of error when we find the columns of $$A$$A one by one via local search.

On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models This paper addresses the general problem of reinforcement learning (RL) in partially observable environments. In 2013, our large RL recurrent neural networks (RNNs) learned from scratch to drive simulated cars from high-dimensional video input. However, real brains are more powerful in many ways. In particular, they learn a predictive model of their initially unknown environment, and somehow use it for abstract (e.g., hierarchical) planning and reasoning. Guided by algorithmic information theory, we describe RNN-based AIs (RNNAIs) designed to do the same. Such an RNNAI can be trained on never-ending sequences of tasks, some of them provided by the user, others invented by the RNNAI itself in a curious, playful fashion, to improve its RNN-based world model. Unlike our previous model-building RNN-based RL machines dating back to 1990, the RNNAI learns to actively query its model for abstract reasoning and planning and decision making, essentially "learning to think." The basic ideas of this report can be applied to many other cases where one RNN-like system exploits the algorithmic information content of another. They are taken from a grant proposal submitted in Fall 2014, and also explain concepts such as "mirror neurons." Experimental results will be described in separate papers.

Deep Learning in Mobile and Wireless Networking: A Survey. The rapid uptake of mobile devices and the rising popularity of mobile applications and services pose unprecedented demands on mobile and wireless networking infrastructure. Upcoming 5G systems are evolving to support exploding mobile traffic volumes, real-time extraction of fine-grained analytics, and agile management of network resources, so as to maximize user experience. Fulfilling these tasks is challenging, as mobile environments are increasingly complex, heterogeneous, and evolving. One potential solution is to resort to advanced machine learning techniques, in order to help manage the rise in data volumes and algorithm-driven applications. The recent success of deep learning underpins new and powerful tools that tackle problems in this space. In this paper we bridge the gap between deep learning and mobile and wireless networking research, by presenting a comprehensive survey of the crossovers between the two areas. We first briefly introduce essential background and state-of-the-art in deep learning techniques with potential applications to networking. We then discuss several techniques and platforms that facilitate the efficient deployment of deep learning onto mobile systems. Subsequently, we provide an encyclopedic review of mobile and wireless networking research based on deep learning, which we categorize by different domains. Drawing from our experience, we discuss how to tailor deep learning to mobile environments. We complete this survey by pinpointing current challenges and open future directions for research.

A Robust Deep Model for Improved Classification of AD/MCI Patients Accurate classification of Alzheimer’s Disease (AD) and its prodromal stage, Mild Cognitive Impairment (MCI), plays a critical role in possibly preventing progression of memory impairment and improving quality of life for AD patients. Among many research tasks, it is of particular interest to identify noninvasive imaging biomarkers for AD diagnosis. In this paper, we present a robust deep learning system to identify different progression stages of AD patients based on MRI and PET scans. We utilized the dropout technique to improve classical deep learning by preventing its weight co-adaptation, which is a typical cause of over-fitting in deep learning. In addition, we incorporated stability selection, an adaptive learning factor, and a multi-task learning strategy into the deep learning framework. We applied the proposed method to the ADNI data set and conducted experiments for AD and MCI conversion diagnosis. Experimental results showed that the dropout technique is very effective in AD diagnosis, improving the classification accuracies by 5.9% on average as compared to the classical deep learning methods.

An Introduction to MCMC for Machine Learning This purpose of this introductory paper is threefold. First, it introduces the Monte Carlo method with emphasis on probabilistic machine learning. Second, it reviews the main building blocks of modern Markov chain Monte Carlo simulation, thereby providing and introduction to the remaining papers of this special issue. Lastly, it discusses new interesting research horizons.

A Simple Weight Decay Can Improve Generalization It has been observed in numerical simulations that a weight decay can im(cid:173) prove generalization in a feed-forward neural network. This paper explains why. It is proven that a weight decay has two effects in a linear network. First, it suppresses any irrelevant components of the weight vector by choosing the smallest vector that solves the learning problem. Second, if the size is chosen right, a weight decay can suppress some of the effects of static noise on the targets, which improves generalization quite a lot. It is then shown how to extend these results to networks with hidden layers and non-linear units. Finally the theory is confirmed by some numerical simulations using the data from NetTalk.

A two-layer ICA-like model estimated by score matching Capturing regularities in high-dimensional data is an important problem in machine learning and signal processing. Here we present a statistical model that learns a nonlinear representation from the data that reflects abstract, invariant properties of the signal without making requirements about the kind of signal that can be processed. The model has a hierarchy of two layers, with the first layer broadly corresponding to Independent Component Analysis (ICA) and a second layer to represent higher order structure. We estimate the model using the mathematical framework of Score Matching (SM), a novel method for the estimation of non-normalized statistical models. The model incorporates a squaring nonlinearity, which we propose to be suitable for forming a higher-order code of invariances. Additionally the squaring can be viewed as modelling subspaces to capture residual dependencies, which linear models cannot capture.

Training Hierarchical Feed-Forward Visual Recognition Models Using Transfer Learning from Pseudo-Tasks Building visual recognition models that adapt across different domains is a challenging task for computer vision. While feature-learning machines in the form of hierarchial feed-forward models (e.g., convolutional neural networks) showed promise in this direction, they are still difficult to train especially when few training examples are available. In this paper, we present a framework for training hierarchical feed-forward models for visual recognition, using transfer learning from pseudo tasks. These pseudo tasks are automatically constructed from data without supervision and comprise a set of simple pattern-matching operations. We show that these pseudo tasks induce an informative inverse-Wishart prior on the functional behavior of the network, offering an effective way to incorporate useful prior knowledge into the network training. In addition to being extremely simple to implement, and adaptable across different domains with little or no extra tuning, our approach achieves promising results on challenging visual recognition tasks, including object recognition, gender recognition, and ethnicity recognition.

Fluents: A Refactoring of Prolog for Uniform Reflection an Interoperation with External Objects On top of a simple kernel (Horn Clause Interpreters with LD-resolution) we introduce Fluents, high level stateful objects which empower and simplify the architecture of logic programming languages through reflection of the underlying interpreter, while providing uniform interoperation patterns with object oriented and procedural languages. We design a Fluent class hierarchy which includes first-class stateful objects representing the meta-level Horn Clause Interpreters, file, URL, socket Readers and Writers, as well as data structures like terms and lists, with high-level operations directly mapped to iterative constructs in the underlying implementation language. Fluents melt naturally in the fabric of Logic Programming languages and provide elegant composition operations, reusability, resource recovery on backtracking and persistence. The Web site of our Kernel Prolog prototype, http://www.binnetcorp.com/kprolog/Main.html allows the reader to try out online the examples discussed in this paper.

DENFIS: dynamic evolving neural-fuzzy inference system and its application for time-series prediction This paper introduces a new type of fuzzy inference systems, denoted as dynamic evolving neural-fuzzy inference system (DENFIS), for adaptive online and offline learning, and their application for dynamic time series prediction. DENFIS evolve through incremental, hybrid (supervised/unsupervised), learning, and accommodate new input data, including new features, new classes, etc., through local element tuning. New fuzzy rules are created and updated during the operation of the system. At each time moment, the output of DENFIS is calculated through a fuzzy inference system based on m-most activated fuzzy rules which are dynamically chosen from a fuzzy rule set. Two approaches are proposed: (1) dynamic creation of a first-order Takagi-Sugeno-type fuzzy rule set for a DENFIS online model; and (2) creation of a first-order Takagi-Sugeno-type fuzzy rule set, or an expanded high-order one, for a DENFIS offline model. A set of fuzzy rules can be inserted into DENFIS before or during its learning process. Fuzzy rules can also be extracted during or after the learning process. An evolving clustering method (ECM), which is employed in both online and offline DENFIS models, is also introduced. It is demonstrated that DENFIS can effectively learn complex temporal sequences in an adaptive way and outperform some well-known, existing models

Optimal prefetching via data compression Caching and prefetching are important mechanisms for speeding up access time to data on secondary storage. Recent work in competitive online algorithms has uncovered several promising new algorithms for caching. In this paper, we apply a form of the competitive philosophy for the first time to the problem of prefetching to develop an optimal universal prefetcher in terms of fault rate, with particular applications to large-scale databases and hypertext systems. Our prediction algorithms with particular applications to large-scale databases and hypertext systems. Our prediction algorithms for prefetching are novel in that they are based on data compression techniques that are both theoretically optimal and good in practice. Intuitively, in order to compress data effectively, you have to be able to predict future data well, and thus good data compressors should be able to predict well for purposes of prefetching. We show for powerful models such as Markov sources and mthe order Markov sources that the page fault rate incurred by our prefetching algorithms are optimal in the limit for almost all sequences of page requests.

Enhancing write I/O performance of disk array RM2 tolerating double disk failures With a large number of internal disks and the rapid growth of disk capacity, storage systems become more susceptible to double disk failures. Thus, the need for such reliable storage systems as RAID6 is expected to gain in importance. However RAID6 architectures such as RM2, P+Q, EVEN-ODD, and DATUM traditionally suffer from a low write I/O performance caused by updating two distinctive parity data associated with user data. To overcome such a low write I/O performance, we propose an enhanced RM2 architecture which combines RM2, one of the well-known RAID6 architectures, with a Lazy Parity Update (LPU) technique. Extensive performance evaluations reveal that the write I/O performance of the proposed architecture is about two times higher than that of RM2 under various I/O workloads with little degradation in reliability.

AirCloud: a cloud-based air-quality monitoring system for everyone We present the design, implementation, and evaluation of AirCloud -- a novel client-cloud system for pervasive and personal air-quality monitoring at low cost. At the frontend, we create two types of Internet-connected particulate matter (PM2:5) monitors -- AQM and miniAQM, with carefully designed mechanical structures for optimal air-flow. On the cloud-side, we create an air-quality analytics engine that learn and create models of air-quality based on a fusion of sensor data. This engine is used to calibrate AQMs and mini-AQMs in real-time, and infer PM2:5 concentrations. We evaluate AirCloud using 5 months of data and 2 month of continuous deployment, and show that AirCloud is able to achieve good accuracies at much lower cost than previous solutions. We also show three real applications built on top of AirCloud by 3rd party developers to further demonstrate the value of our system.

The size of MDP factored policies Policies of Markov Decision Processes (MDPs) tell the next action to execute, given the current state and (possibly) the history of actions executed so far. Factorization is used when the number of states is exponentially large: both the MDP and the policy can be then represented using a compact form, for example employing circuits. We prove that there are MDPs whose optimal policies require exponential space evenin factored form.

Monotonic reductions, representative equivalence, and compilation of intractable problems The idea of preprocessing part of the input of a problem in order to improve efficiency has been employed by several researchers in several areas of computer science. In this article, we show sufficient conditions to prove that an intractable problem cannot be efficiently solved even allowing an exponentially long preprocessing phase. The generality of such conditions is shown by applying them to various problems coming from different fields. While the results may seem to discourage the use of compilation, we present some evidence that such negative results are useful in practice.

On the undecidability of probabilistic planning and infinite-horizon partially observable Markov decision problems We investigate the computability of problems in probabilistic planning and partially observable infinite-horizon Markov decision processes. The undecidability of the string-existence problem for probabilistic finite automata is adapted to show that the following problem of plan existence in probabilistic planning is undecidable: given a probabilistic planning problem, determine whether there exists a plan with success probability exceeding a desirable threshold. Analogous policy-existence problems for partially observable infinite-horizon Markov decision processes under discounted and undiscounted total reward models, average-reward models, and state-avoidance models are all shown to be undecidable. The results apply to corresponding approximation problems as well.

Complexity of finite-horizon Markov decision process problems Controlled stochastic systems occur in science engineering, manufacturing, social sciences, and many other cntexts. If the systems is modeled as a Markov decision process (MDP) and will run ad infinitum, the optimal control policy can be computed in polynomial time using linear programming. The problems considered here assume that the time that the process will run is finite, and based on the size of the input. There are mny factors that compound the complexity of computing the optimal policy. For instance, there are many factors that compound the complexity of this computation. For instance, if the controller does not have complete information about the state of the system, or if the system is represented in some very succint manner, the optimal policy is provably not computable in time polynomial in the size of the input. We analyze the computational complexity of evaluating policies and of determining whether a sufficiently good policy exists for a MDP, based on a number of confounding factors, including the observability of the system state; the succinctness of the representation; the type of policy; even the number of actions relative to the number of states. In almost every case, we show that the decision problem is complete for some known complexity class. Some of these results are familiar from work by Papadimitriou and Tsitsiklis and others, but some, such as our PL-completeness proofs, are surprising. We include proofs of completeness for natural problems in the as yet little-studied classes NPPP.

On the Compilability of Diagnosis, Planning, Reasoning about Actions, Belief Revision, etc In this paper we investigate the usefulness of preprocessing part of the input of a given problem to improve the efficiency. We extend the results of (Cadoli et al., 1996) by giving sufficient conditions to prove the unfeasibility of reducing the on-line complexity via an off- line preprocessing. We analyze the problems of diagnosis (Peng and Reggia, 1986), plan- ning (Bylander, 1991), reasoning about ac- tions (Gelfond and Lifschitz, 1993), and be- lief revision (Williams, 1994). We analyze other problems from various fields.

Exploiting Structure in Policy Construction Markov decision processes (MDPs) have recently been applied to the problem of modeling decision-theoretic planning. While traditional methods for solving MDPs are often practical for small states spaces, their effectiveness for large AI planning problems is questionable. We present an algorithm, called structured policy Iteration (SPI), that constructs optimal policies without explicit enumeration of the state space. The algorithm retains the fundamental computational steps of the commonly used modified policy iteration algorithm, but exploits the variable and prepositional independencies reflected in a temporal Bayesian network representation of MDPs. The principles behind SPI can be applied to any structured representation of stochastic actions, policies and value functions, and the algorithm itself can be used in conjunction with recent approximation methods.

Relationships between nondeterministic and deterministic tape complexities The amount of storage needed to simulate a nondeterministic tape bounded Turingmachine on a deterministic Turing machine is investigated. Results include the following: Theorem. A nondeterministic L(n)-tape bounded Turing machine can be simulated by a deterministic [L(n)]^2-tape bounded Turing machine, provided L(n)=log"2n. Computations of nondeterministic machines are shown to correspond to threadings of certain mazes. This correspondence is used to produce a specific set, namely the set of all codings of threadable mazes, such that, if there is any set which distinguishes nondeterministic tape complexity classes from deterministic tape complexity classes, then this is one such set.

Reasoning about partially ordered events This paper describes a class of temporal reasoning problems involving events whose order is not completely known. We examine the complexity of such problems and show that for all but trivial cases these problems are likely to be intractable. As an alternative to a complete, but potentially exponential-time decision procedure, we provide a partial decision procedure that reports useful results and runs in polynomial time.

Counting, Selecting, adn Sorting by Query-Bounded Machines We study the query-complexity of counting, selecting, and sorting functions. That is, for a given set A and a positive integer k, we ask, how many queries to an arbitrary oracle does a polynomial-time machine on input (x 1, x 2,..., x k ) need to determine how many strings of the input are in A. We also ask how many queries are necessary to select a string in A from the input (x 1, x 2,..., x k ) if such a string exists and to sort the input (x 1, x 2,..., x k ) with respect to the ordering x y if and only if x A y A. We obtain optimal query-bounds for these problems, and show that sets for which these functions have a low query-complexity must be easy in some sense. For such sets we obtain optimal placements in the extended low hierarchy. We also show that in the case of NP-complete sets the lower bounds for counting and selecting hold unless P=NP. Finally, we relate these notions to cheatability and p-superterseness. Our results yield as corollaries extensions of previously know results.

Stacked generalization This paper introduces stacked generalization, a scheme for minimizing the generalization error rate of one or more generalizers. Stacked generalization works by deducing the biases of the generalizer(s) with respect to a provided learning set. This deduction proceeds by generalizing in a second space whose inputs are (for example) the guesses of the original generalizers when taught with part of the learning set and trying to guess the rest of it, and whose output is (for example) the correct guess. When used with multiple generalizers, stacked generalization can be seen as a more sophisticated version of cross-validation, exploiting a strategy more sophisticated than cross-validation's crude winner-takes-all for combining the individual generalizers. When used with a single generalizer, stacked generalization is a scheme for estimating (and then correcting for) the error of a generalizer which has been trained on a particular learning set and then asked a particular question. After introducing stacked generalization and justifying its use, this paper presents two numerical experiments. The first demonstrates how stacked generalization improves upon a set of separate generalizers for the NETtalk task of translating text to phonemes. The second demonstrates how stacked generalization improves the performance of a single surface-fitter. With the other experimental evidence in the literature, the usual arguments supporting cross-validation, and the abstract justifications presented in this paper, the conclusion is that for almost any real-world generalization problem one should use some version of stacked generalization to minimize the generalization error rate. This paper ends by discussing some of the variations of stacked generalization, and how it touches on other fields like chaos theory.

The role of macros in tractable planning This paper presents several new tractability results for planning based on macros. We describe an algorithm that optimally solves planning problems in a class that we call inverted tree reducible, and is provably tractable for several subclasses of this class. By using macros to store partial plans that recur frequently in the solution, the algorithm is polynomial in time and space even for exponentially long plans. We generalize the inverted tree reducible class in several ways and describe modifications of the algorithm to deal with these new classes. Theoretical results are validated in experiments.

Higher order contractive auto-encoder We propose a novel regularizer when training an autoencoder for unsupervised feature extraction. We explicitly encourage the latent representation to contract the input space by regularizing the norm of the Jacobian (analytically) and the Hessian (stochastically) of the encoder's output with respect to its input, at the training points. While the penalty on the Jacobian's norm ensures robustness to tiny corruption of samples in the input space, constraining the norm of the Hessian extends this robustness when moving further away from the sample. From a manifold learning perspective, balancing this regularization with the auto-encoder's reconstruction objective yields a representation that varies most when moving along the data manifold in input space, and is most insensitive in directions orthogonal to the manifold. The second order regularization, using the Hessian, penalizes curvature, and thus favors smooth manifold. We show that our proposed technique, while remaining computationally efficient, yields representations that are significantly better suited for initializing deep architectures than previously proposed approaches, beating state-of-the-art performance on a number of datasets.

MAXSAT Heuristics for Cost Optimal Planning.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Pre-training of Recurrent Neural Networks via Linear Autoencoders. We propose a pre-training technique for recurrent neural networks based on linear autoencoder networks for sequences, i.e. linear dynamical systems modelling the target sequences. We start by giving a closed form solution for the definition of the optimal weights of a linear autoencoder given a training set of sequences. This solution, however, is computationally very demanding, so we suggest a procedure to get an approximate solution for a given number of hidden units. The weights obtained for the linear autoencoder are then used as initial weights for the input-to-hidden connections of a recurrent neural network, which is then trained on the desired task. Using four well known datasets of sequences of polyphonic music, we show that the proposed pre-training approach is highly effective, since it allows to largely improve the state of the art results on all the considered datasets.

Three new graphical models for statistical language modelling The supremacy of n-gram models in statistical language modelling has recently been challenged by parametric models that use distributed representations to counteract the difficulties caused by data sparsity. We propose three new probabilistic language models that define the distribution of the next word in a sequence given several preceding words by using distributed representations of those words. We show how real-valued distributed representations for words can be learned at the same time as learning a large set of stochastic binary hidden features that are used to predict the distributed representation of the next word from previous distributed representations. Adding connections from the previous states of the binary hidden features improves performance as does adding direct connections between the real-valued distributed representations. One of our models significantly outperforms the very best n-gram models.

Modeling Temporal Dependencies in High-Dimensional Sequences: Application to Polyphonic Music Generation and Transcription. We investigate the problem of modeling symbolic sequences of polyphonic music in a completely general piano-roll representation. We introduce a probabilistic model based on distribution estimators conditioned on a recurrent neural network that is able to discover temporal dependencies in high-dimensional sequences. Our approach outperforms many traditional models of polyphonic music on a variety of realistic datasets. We show how our musical language model can serve as a symbolic prior to improve the accuracy of polyphonic transcription.

A neural probabilistic language model A goal of statistical language modeling is to learn the joint probability function of sequences of words in a language. This is intrinsically difficult because of the curse of dimensionality: a word sequence on which the model will be tested is likely to be different from all the word sequences seen during training. Traditional but very successful approaches based on n-grams obtain generalization by concatenating very short overlapping sequences seen in the training set. We propose to fight the curse of dimensionality by learning a distributed representation for words which allows each training sentence to inform the model about an exponential number of semantically neighboring sentences. The model learns simultaneously (1) a distributed representation for each word along with (2) the probability function for word sequences, expressed in terms of these representations. Generalization is obtained because a sequence of words that has never been seen before gets high probability if it is made of words that are similar (in the sense of having a nearby representation) to words forming an already seen sentence. Training such large models (with millions of parameters) within a reasonable time is itself a significant challenge. We report on experiments using neural networks for the probability function, showing on two text corpora that the proposed approach significantly improves on state-of-the-art n-gram models, and that the proposed approach allows to take advantage of longer contexts.

Learning Deep Architectures for AI Theoretical results suggest that in order to learn the kind of com-plicated functions that can represent high-level abstractions (e.g., invision, language, and other AI-level tasks), one may needdeep architec-tures. Deep architectures are composed of multiple levels of non-linearoperations, such as in neural nets with many hidden layers or in com-plicated propositional formulae re-using many sub-formulae. Searchingthe parameter space of deep architectures is a difficult task, but learningalgorithms such as those for Deep Belief Networks have recently beenproposed to tackle this problem with notable success, beating the state-of-the-art in certain areas. This monograph discusses the motivationsand principles regarding learning algorithms for deep architectures, inparticular those exploiting as building blocks unsupervised learning ofsingle-layer models such as Restricted Boltzmann Machines, used toconstruct deeper models such as Deep Belief Networks.

Self Supervised Boosting Boosting algorithms and successful applications thereof abound for clas- sification and regression learning problems, but not for unsupervised learning. We propose a sequential approach to adding features to a ran- dom field model by training them to improve classification performance between the data and an equal-sized sample of "negative examples" gen- erated from the model's current estimate of the data density. Training in each boosting round proceeds in three stages: first we sample negative examples from the model's current Boltzmann distribution. Next, a fea- ture is trained to improve classification performance between data and negative examples. Finally, a coefficient is learned which determines the importance of this feature relative to ones already in the pool. Negative examples only need to be generated once to learn each new feature. The validity of the approach is demonstrated on binary digits and continuous synthetic data.

A Nonparametric Bayesian Approach to Modeling Overlapping Clusters Although clustering data into mutually ex- clusive partitions has been an extremely suc- cessful approach to unsupervised learning, there are many situations in which a richer model is needed to fully represent the data. This is the case in problems where data points actually simultaneously belong to mul- tiple, overlapping clusters. For example a particular gene may have several functions, therefore belonging to several distinct clus- ters of genes, and a biologist may want to discover these through unsupervised model- ing of gene expression data. We present a new nonparametric Bayesian method, the In- finite Overlapping Mixture Model (IOMM), for modeling overlapping clusters. The IOMM uses exponential family distributions to model each cluster and forms an over- lapping mixture by taking products of such distributions, much like products of experts (Hinton, 2002). The IOMM allows an un- bounded number of clusters, and assignments of points to (multiple) clusters is modeled us- ing an Indian Buet Process (IBP), (Griths and Ghahramani, 2006). The IOMM has the desirable properties of being able to focus in on overlapping regions while maintaining the ability to model a potentially infinite num- ber of clusters which may overlap. We derive MCMC inference algorithms for the IOMM and show that these can be used to cluster movies into multiple genres.

Compressed Sensing. Suppose x is an unknown vector in Ropfm (a digital image or signal); we plan to measure n general linear functionals of x and then reconstruct. If x is known to be compressible by transform coding with a known transform, and we reconstruct via the nonlinear procedure defined here, the number of measurements n can be dramatically smaller than the size m. Thus, certain natural classes of images with m pixels need only n=O(m1/4log5/2(m)) nonadaptive nonpixel samples for faithful recovery, as opposed to the usual m pixel samples. More specifically, suppose x has a sparse representation in some orthonormal basis (e.g., wavelet, Fourier) or tight frame (e.g., curvelet, Gabor)-so the coefficients belong to an lscrp ball for 0<ples1. The N most important coefficients in that expansion allow reconstruction with lscr2 error O(N1/2-1p/). It is possible to design n=O(Nlog(m)) nonadaptive measurements allowing reconstruction with accuracy comparable to that attainable with direct knowledge of the N most important coefficients. Moreover, a good approximation to those N important coefficients is extracted from the n measurements by solving a linear program-Basis Pursuit in signal processing. The nonadaptive measurements have the character of "random" linear combinations of basis/frame elements. Our results use the notions of optimal recovery, of n-widths, and information-based complexity. We estimate the Gel'fand n-widths of lscrp balls in high-dimensional Euclidean space in the case 0<ples1, and give a criterion identifying near- optimal subspaces for Gel'fand n-widths. We show that "most" subspaces are near-optimal, and show that convex optimization (Basis Pursuit) is a near-optimal way to extract information derived from these near-optimal subspaces

Training restricted Boltzmann machines using approximations to the likelihood gradient A new algorithm for training Restricted Boltzmann Machines is introduced. The algorithm, named Persistent Contrastive Divergence, is different from the standard Contrastive Divergence algorithms in that it aims to draw samples from almost exactly the model distribution. It is compared to some standard Contrastive Divergence and Pseudo-Likelihood algorithms on the tasks of modeling and classifying various types of data. The Persistent Contrastive Divergence algorithm outperforms the other algorithms, and is equally fast and simple.

Well founded semantics for logic programs with explicit negation . The aim of this paper is to provide asemantics for general logic programs (with negation bydefault) extended with explicit negation, subsumingwell founded semantics [22].The Well Founded semantics for extended logicprograms (WFSX) is expressible by a default theorysemantics we have devised [11]. This relationshipimproves the cross--fertilization between logic programsand default theories, since we generalize previousresults concerning their relationship [3, 4, 7, 1, 2],and there is...

An overview of MetaMap: historical perspective and recent advances. MetaMap is a widely available program providing access to the concepts in the unified medical language system (UMLS) Metathesaurus from biomedical text. This study reports on MetaMap's evolution over more than a decade, concentrating on those features arising out of the research needs of the biomedical informatics community both within and outside of the National Library of Medicine. Such features include the detection of author-defined acronyms/abbreviations, the ability to browse the Metathesaurus for concepts even tenuously related to input text, the detection of negation in situations in which the polarity of predications is important, word sense disambiguation (WSD), and various technical and algorithmic features. Near-term plans for MetaMap development include the incorporation of chemical name recognition and enhanced WSD.

A fully associative software-managed cache design As DRAM access latencies approach a thousand instruction-execution times and on-chip caches grow to multiple megabytes, it is not clear that conventional cache structures continue to be appropriate. Two key features—full associativity and software management—have been used successfully in the virtual-memory domain to cope with disk access latencies. Future systems will need to employ similar techniques to deal with DRAM latencies. This paper presents a practical, fully associative, software-managed secondary cache system that provides performance competitive with or superior to traditional caches without OS or application involvement. We see this structure as the first step toward OS- and application-aware management of large on-chip caches.This paper has two primary contributions: a practical design for a fully associative memory structure, the indirect index cache (IIC), and a novel replacement algorithm, generational replacement, that is specifically designed to work with the IIC. We analyze the behavior of an IIC with generational replacement as a drop-in, transparent substitute for a conventional secondary cache. We achieve miss rate reductions from 8% to 85% relative to a 4-way associative LRU organization, matching or beating a (practically infeasible) fully associative true LRU cache. Incorporating these miss rates into a rudimentary timing model indicates that the IIC/generational replacement cache could be competitive with a conventional cache at today's DRAM latencies, and will outperform a conventional cache as these CPU-relative latencies grow.

Towards application/file-level characterization of block references: a case for fine-grained buffer management Two contributions are made in this paper. First, we show that system level characterization of file block references is inadequate for maximizing buffer cache performance. We show that a finer-grained characterization approach is needed. Though application level characterization methods have been proposed, this is the first attempt, to the best of our knowledge, to consider file level characterizations. We propose an Application/File-level Characterization (AFC) scheme where we detect on-line the reference characteristics at the application level and then at the file level, if necessary. The results of this characterization are used to employ appropriate replacement policies in the buffer cache to maximize performance. The second contribution is in proposing an efficient and fair buffer allocation scheme. Application or file level resource management is infeasible unless there exists an allocation scheme that is efficient and fair. We propose the &Dgr;HIT allocation scheme that takes away a block from the application/file where the removal results in the smallest reduction in the number of expected buffer cache hits. Both the AFC and &Dgr;HIT schemes are on-line schemes that detect and allocate as applications execute. Experiments using trace-driven simulations show that substantial performance improvements can be made. For single application executions the hit ratio increased an average of 13 percentage points compared to the LRU policy, with a maximum increase of 59 percentage points, while for multiple application executions, the increase is an average of 12 percentage points, with a maximum of 32 percentage points for the workloads considered.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Working Set Size Estimation Techniques in Virtualized Environments: One Size Does not Fit All. Energy consumption is a primary concern for datacenters' management. Numerous datacenters are relying on virtualization, as it provides flexible resource management means such as virtual machine (VM) checkpoint/restart, migration and consolidation. However, one of the main hindrances to server consolidation is physical memory. In nowadays cloud, memory is generally statically allocated to VMs and wasted if not used. Techniques (such as ballooning) were introduced for dynamically reclaiming memory from VMs, such that only the needed memory is provisioned to each VM. However, the challenge is to precisely monitor the needed memory, i.e., the working set of each VM. In this paper, we thoroughly review the main techniques that were proposed for monitoring the working set of VMs. Additionally, we have implemented the main techniques in the Xen hypervisor and we have defined different metrics in order to evaluate their efficiency. Based on the evaluation results, we propose Badis, a system which combines several of the existing solutions, using the right solution at the right time. We also propose a consolidation extension which leverages Badis in order to pack the VMs based on the working set size and not the booked memory.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Extending Group Communication Facilities To Support Complex Distributed Office Procedures Complex distributed office procedures are based on sequential and parallel execution of basic services offered by a distributed service environment. Office procedure execution can be considered to be a special kind of group communication with groups of collaborating servers. This paper first presents a general group communication abstraction. We then show in detail how the generic approach has been extended and specialized in order to support complex office procedures. The extended solution comprises a graph-based, declarative specification notation for office procedures, and a distributed runtime support environment. This maps office procedure service requests dynamically to actual servers for execution. Moreover, additional management facilities enable external supervision of office procedures. Based on examples, we show how applications can benefit from the abstractions. By discussing their similarities and evolution, we also outline the strong relationship between group communication and office procedures.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

A Syntax-Based Approach to Reasoning about Actions and Events In this paper, we introduce an alternative approach to reasoning about action. The approach provides a solution to the frame and the ramification problem in a uniform manner. The approach involves keeping a (syntax-based) model of the world that is updated when actions are performed. The approach is similar to the STRIPS system in which formulas are deleted and added as effects of an action. The approach however does not suffer from STRIPS' limitations in expressivity.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Word Alignment Modeling with Context Dependent Deep Neural Network.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Multi-source deep learning for information trustworthiness estimation In recent years, information trustworthiness has become a serious issue when user-generated contents prevail in our information world. In this paper, we investigate the important problem of estimating information trustworthiness from the perspective of correlating and comparing multiple data sources. To a certain extent, the consistency degree is an indicator of information reliability--Information unanimously agreed by all the sources is more likely to be reliable. Based on this principle, we develop an effective computational approach to identify consistent information from multiple data sources. Particularly, we analyze vast amounts of information collected from multiple review platforms (multiple sources) in which people can rate and review the items they have purchased. The major challenge is that different platforms attract diverse sets of users, and thus information cannot be compared directly at the surface. However, latent reasons hidden in user ratings are mostly shared by multiple sources, and thus inconsistency about an item only appears when some source provides ratings deviating from the common latent reasons. Therefore, we propose a novel two-step procedure to calculate information consistency degrees for a set of items which are rated by multiple sets of users on different platforms. We first build a Multi-Source Deep Belief Network (MSDBN) to identify the common reasons hidden in multi-source rating data, and then calculate a consistency score for each item by comparing individual sources with the reconstructed data derived from the latent reasons. We conduct experiments on real user ratings collected from Orbitz, Priceline and TripAdvisor on all the hotels in Las Vegas and New York City. Experimental results demonstrate that the proposed approach successfully finds the hotels that receive inconsistent, and possibly unreliable, ratings.

Restricted deep belief networks for multi-view learning Deep belief network (DBN) is a probabilistic generative model with multiple layers of hidden nodes and a layer of visible nodes, where parameterizations between layers obey harmonium or restricted Boltzmann machines (RBMs). In this paper we present restricted deep belief network (RDBN) for multi-view learning, where each layer of hidden nodes is composed of view-specific and shared hidden nodes, in order to learn individual and shared hidden spaces from multiple views of data. View-specific hidden nodes are connected to corresponding view-specific hidden nodes in the lower-layer or visible nodes involving a specific view, whereas shared hidden nodes follow inter-layer connections without restrictions as in standard DBNs. RDBN is trained using layer-wise contrastive divergence learning. Numerical experiments on synthetic and real-world datasets demonstrate the useful behavior of the RDBN, compared to the multi-wing harmonium (MWH) which is a two-layer undirected model.

A neural probabilistic language model A goal of statistical language modeling is to learn the joint probability function of sequences of words in a language. This is intrinsically difficult because of the curse of dimensionality: a word sequence on which the model will be tested is likely to be different from all the word sequences seen during training. Traditional but very successful approaches based on n-grams obtain generalization by concatenating very short overlapping sequences seen in the training set. We propose to fight the curse of dimensionality by learning a distributed representation for words which allows each training sentence to inform the model about an exponential number of semantically neighboring sentences. The model learns simultaneously (1) a distributed representation for each word along with (2) the probability function for word sequences, expressed in terms of these representations. Generalization is obtained because a sequence of words that has never been seen before gets high probability if it is made of words that are similar (in the sense of having a nearby representation) to words forming an already seen sentence. Training such large models (with millions of parameters) within a reasonable time is itself a significant challenge. We report on experiments using neural networks for the probability function, showing on two text corpora that the proposed approach significantly improves on state-of-the-art n-gram models, and that the proposed approach allows to take advantage of longer contexts.

A fast learning algorithm for deep belief nets. We show how to use "complementary priors" to eliminate the explaining-away effects that make inference difficult in densely connected belief nets that have many hidden layers. Using complementary priors, we derive a fast, greedy algorithm that can learn deep, directed belief networks one layer at a time, provided the top two layers form an undirected associative memory. The fast, greedy algorithm is used to initialize a slower learning procedure that fine-tunes the weights using a contrastive version of the wake-sleep algorithm. After fine-tuning, a network with three hidden layers forms a very good generative model of the joint distribution of handwritten digit images and their labels. This generative model gives better digit classification than the best discriminative learning algorithms. The low-dimensional manifolds on which the digits lie are modeled by long ravines in the free-energy landscape of the top-level associative memory, and it is easy to explore these ravines by using the directed connections to display what the associative memory has in mind.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

On the facial structure of set packing polyhedra In this paper we address ourselves to identifying facets of the set packing polyhedron, i.e., of the convex hull of integer solutions to the set covering problem with equality constraints and/or constraints of the form “?”. This is done by using the equivalent node-packing problem derived from the intersection graph associated with the problem under consideration. First, we show that the cliques of the intersection graph provide a first set of facets for the polyhedron in question. Second, it is shown that the cycles without chords of odd length of the intersection graph give rise to a further set of facets. A rather strong geometric property of this set of facets is exhibited.

Proceedings of the 17th ACM Conference on Information and Knowledge Management, CIKM 2008, Napa Valley, California, USA, October 26-30, 2008

Monotonic reductions, representative equivalence, and compilation of intractable problems The idea of preprocessing part of the input of a problem in order to improve efficiency has been employed by several researchers in several areas of computer science. In this article, we show sufficient conditions to prove that an intractable problem cannot be efficiently solved even allowing an exponentially long preprocessing phase. The generality of such conditions is shown by applying them to various problems coming from different fields. While the results may seem to discourage the use of compilation, we present some evidence that such negative results are useful in practice.

Comparative Evaluation of Latency Tolerance Techniques for Software Distributed Shared Memory A key challenge in achieving high performance on software DSMs is overcoming their relatively large communication latencies. In this paper, we consider two techniques which address this problem: prefetching and multithreading. While previous studies have examined each of these techniques in isolation, this paper is the first to evaluate both techniques using a consistent hardware platform and set of applications, thereby allowing direct comparisons. In addition, this is the first study to consider combining prefetching and multithreading in a software DSM. We performed our experiments on real hardware using a full implementation of both techniques. Our experimental results demonstrate that both prefetching and multithreading result in significant performance improvements when applied individually. In addition, we observe that prefetching and multithreading can potentially complement each other by using prefetching to hide memory latency and multithreading to hide synchronization latency.

Reasoning About Actions in Narrative Understanding Reasoning about actions has been a focus of interest in AI from the beginning and continues to receive attention. Rut the range of situations considered has been rather narrow and falls well short of what is needed for understanding natural language. Language understanding requires sophisticated reasoning about actions and events and the world's languages employ a variety of grammatical and lexical devices to construe, direct attention and focus on, and control inferences about actions and events. We implemented a neurally inspired computational model that is able to reason about, linguistic action and event descriptions, such as those found in news stories. The system uses an active. event representation that also seems to provide natural and cognitiveIy motivated solutions to classical problems in logical theories of reasoning about actions. For logical approaches to reasoning about actions, we suggest, that looking at story understanding sets up fairly strong desiderata both in terms of the fine-grained event and action distinctions and the kinds of real-time inferences required.

Small cache, big effect: provable load balancing for randomly partitioned cluster services Load balancing requests across a cluster of back-end servers is critical for avoiding performance bottlenecks and meeting service-level objectives (SLOs) in large-scale cloud computing services. This paper shows how a small, fast popularity-based front-end cache can ensure load balancing for an important class of such services; furthermore, we prove an O(n log n) lower-bound on the necessary cache size and show that this size depends only on the total number of back-end nodes n, not the number of items stored in the system. We validate our analysis through simulation and empirical results running a key-value storage system on an 85-node cluster.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Error Control for the Localized Reduced Basis Multiscale Method with Adaptive On-Line Enrichment. In this contribution we consider localized, robust, and efficient a posteriori error estimation of the localized reduced basis multiscale (LRBMS) method for parametric elliptic problems with possibly heterogeneous diffusion coefficient. The numerical treatment of such parametric multiscale problems is characterized by a high computational complexity, arising from the multiscale character of the underlying differential equation and the additional parameter dependence. The LRBMS method can be seen as a combination of numerical multiscale methods and model reduction using reduced basis (RB) methods to efficiently reduce the computational complexity with respect to the multiscale as well as the parametric aspect of the problem, simultaneously. In contrast to the classical residual based error estimators currently used in RB methods, we are considering error estimators that are based on conservative flux reconstruction and provide an efficient and rigorous bound on the full error with respect to the weak solution. In addition, the resulting error estimator is localized and can thus be used in the on-line phase to adaptively enrich the solution space locally where needed. The resulting certified LRBMS method with adaptive on-line enrichment thus guarantees the quality of the reduced solution during the on-line phase, given any (possibly insufficient) reduced basis that was generated during the off-line phase. Numerical experiments are given to demonstrate the applicability of the resulting algorithm with on-line enrichment to single phase flow in heterogeneous media.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Simultaneous Strong Separations of Probabilistic and Unambiguous Complexity Classes We study the relationship between probabilistic and unambiguous computation, and provide strong relativized evidence that they are incomparable. In particular, we display a relativized world in which the complexity classes embodying these paradigms of computation are mutually immune. We answer questions formulated in|and extend the line of research opened by|Geske and Grollman (15) and Balcazar and Russo (3).

The random oracle hypothesis is false In this paper, P(#P) and PF(#P) are characterized in terms of a largely different computation structure, where P(#P) (resp., PF(#P)) is the class of sets (resp., functions) that are polynomial-time Turing reducible to #P functions. Let MidP be the class ...

Banishing Robust Turing Completeness ABSTRACT This paper proves that \promise classes" are so fragilely structured that they do not robustly (i.e. with respect to all oracles) possess Turinghard sets even in classes far larger than themselves. In particular, this paper shows that FewP does not robustly possess Turing hard sets for UP \ coUP and IP \ coIP does not robustly possess Turing hard sets for ZPP. It follows that ZPP, R, coR, UP\coUP, UP, FewP\coFewP, FewP, and IP \ coIP do not robustly possess Turing complete sets. This both resolves open questions of whether promise classes lacking robust downward closure under Turing reductions (e.g., R, UP, FewP) might robustly have Turing complete sets, and extends the range of classes known not to robustly contain many-one complete sets. Keywords: Structural complexity theory; Polynomial-time reductions;

Two remarks on the power of counting The relationship between the polynomial hierarchy and Valiant's class #P is at present unknown. We show that some low portions of the polynomial hierarchy, namely deterministic polynomial algorithms using an NP oracle at most a logarithmic number of times, can be simulated by one #P computation. We also show that the class of problems solvable by polynomial-time nondeterministic Turing machines which accept whenever there is an odd number of accepting computations is idempotent, that is, closed under usage of oracles from the same class.

Nondeterministic turing machines with modified acceptance

A Downward Collapse within the Polynomial Hierarchy Downward collapse (also known as upward separation) refers to cases where the equality of two larger classes implies the equality of two smaller classes. We provide an unqualified downward collapse result completely within the polynomial hierarchy. In particular, we prove that, for k 2, if ${\rm P}^{\Sigma^p_k[1]} = {\rm P}^{\Sigma^p_k[2]}$ then $\Sigma^p_k = \Pi^p_k = {\rm PH}$. We extend this to obtain a more general downward collapse result.

On boolean lowness and boolean highness The concepts of lowness and highness originate from recursion theory and were introduced into the complexity theory by Schoning (Lecture Notes in Computer Science, Vol. 211, Springer, Berlin, 1985). Informally, a set is low (high resp.) for a relativizable class K of languages if it does not add (adds maximal resp.) power to K when used as an oracle. In this paper, we introduce the notions of boolean lowness and boolean highness. Informally, a set is boolean low (boolean high resp.) for a class X of languages if it does not add (adds maximal resp.) power to K when combined with K by boolean operations. We prove properties of boolean lowness and boolean highness which show a lot of similarities with the notions of lowness and highness. Using Kadin's technique of hard strings (see Kadin, SIAM J. Comput 17(6) (1988) 1263-1282; Wagner, Number-of-query hierachies, TR 158, University of Augsburg, 1987; Chang and Kadin SIAM J. Comput. 25(2) (1996) 340; Beigel ct al. Math. Systems Theory 26 (1993) 293-310) we show that the sets which are boolean low for the classes of the boolean hierarchy are low for the boolean closure of Sigma (p)(2). Furthermore, we prove a result on boolean lowness which has as a corollary the best known result (sec Beigel, (1993); in fact even a bit better) on the connection of the collapses of the boolean hierarchy and the polynomial-lime hierarchy if BH = NP(L) then PH = Sigma (p)(2)(k - 1) circle plus NP(k). (C) 2001 Published by Elsevier Science B.V.

The complexity of facets (and some facets of complexity) Many important combinatorial optimization problems, including the traveling salesman problem (TSP), the clique problem and many others, call for the optimization of a linear functional over some discrete set of vectors.

Evaluating collaborative filtering recommender systems Recommender systems have been evaluated in many, often incomparable, ways. In this article, we review the key decisions in evaluating collaborative filtering recommender systems: the user tasks being evaluated, the types of analysis and datasets being used, the ways in which prediction quality is measured, the evaluation of prediction attributes other than quality, and the user-based evaluation of the system as a whole. In addition to reviewing the evaluation strategies used by prior researchers, we present empirical results from the analysis of various accuracy metrics on one content domain where all the tested metrics collapsed roughly into three equivalence classes. Metrics within each equivalency class were strongly correlated, while metrics from different equivalency classes were uncorrelated.

Adaptive page replacement based on memory reference behavior As disk performance continues to lag behind that of memory systems and processors, virtual memory management becomes increasingly important for overall system performance. In this paper we study the page reference behavior of a collection of memory-intensive applications, and propose a new virtual memory page replacement algorithm, SEQ. SEQ detects long sequences of page faults and applies most-recently-used replacement to those sequences. Simulations show that for a large class of applications, SEQ performs close to the optimal replacement algorithm, and significantly better than Least-Recently-Used (LRU). In addition, SEQ performs similarly to LRU for applications that do not exhibit sequential faulting.

Logic programs with stable model semantics as a constraint programming paradigm Logic programming with the stable model semantics is put forward as a novel constraint programming paradigm. This paradigm is interesting because it bring advantages of logic programming based knowledge representation techniques to constraint programming and because implementation methods for the stable model semantics for ground (variable&dash;free) programs have advanced significantly in recent years. For a program with variables these methods need a grounding procedure for generating a variable&dash;free program. As a practical approach to handling the grounding problem a subclass of logic programs, domain restricted programs, is proposed. This subclass enables efficient grounding procedures and serves as a basis for integrating built&dash;in predicates and functions often needed in applications. It is shown that the novel paradigm embeds classical logical satisfiability and standard (finite domain) constraint satisfaction problems but seems to provide a more expressive framework from a knowledge representation point of view. The first steps towards a programming methodology for the new paradigm are taken by presenting solutions to standard constraint satisfaction problems, combinatorial graph problems and planning problems. An efficient implementation of the paradigm based on domain restricted programs has been developed. This is an extension of a previous implementation of the stable model semantics, the Smodels system, and is publicly available. It contains, e.g., built&dash;in integer arithmetic integrated to stable model computation. The implementation is described briefly and some test results illustrating the current level of performance are reported.

Reducing file system latency using a predictive approach Despite impressive advances in file system through put resulting from technologies such as high-bandwidth networks and disk arrays, file system latency has not improved and in many cases has become worse. Consequently, file system I/O remains one of the major bottlenecks to operating system performance [10]. This paper investigates an automated predictive approach towards reducing file latency. Automatic Prefetching uses past file accesses to predict future file systemrequests. The objective is to provide data in advance of the request for the data, effectively masking access latencies. We have designed and implement a system to measure the performance benefits of automatic prefetching. Our current results, obtained from a trace-driven simulation, show that prefetching results in as much as a 280% improvement over LRU especially for smaller caches. Alternatively, prefetching can reduce cache size by up to 50%.

Global reinforcement learning in neural networks. In this letter, we have found a more general formulation of the REward Increment = Nonnegative Factor x Offset Reinforcement x Characteristic Eligibility (REINFORCE) learning principle first suggested by Williams. The new formulation has enabled us to apply the principle to global reinforcement learning in networks with various sources of randomness, and to suggest several simple local rules for such networks. Numerical simulations have shown that for simple classification and reinforcement learning tasks, at least one family of the new learning rules gives results comparable to those provided by the famous Rules A(r-i) and A(r-p) for the Boltzmann machines.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Write twice disk buffering Non-volatile (safe) disk buffers improve performance in DBMSs. Writes to the disk buffer are durable without incurring physical disk writes. Moreover, physical writes at the disk can be performed at low cost using write piggybacking and multiblock purges. However, safe disk buffers are expensive.In this paper, unsafe buffers are used to simulate safe buffers, by writing each block twice to disk. The first write is performed immediately, anywhere. The second write is deferred, but written at a fixed location. These two writes together cost less than a single random write.

Dynamic Data Distribution (D3) in a Shared-Nothing Multiprocessor Data Store

Incremental recovery in main memory database systems Recovery activities, like checkpointing and restart, in traditional database management systems are performed in a quiescent state where no transactions are active. This approach impairs the performance of online transaction processing systems, especially when a large volatile memory is used. An incremental scheme for performing recovery in main memory database systems (MMDBs), in parallel with transaction execution, is presented. A page-based incremental restart algorithm that enables the resumption of transaction processing as soon as the system is up is proposed. Pages are recovered individually and according to the demands of the post-crash transactions. A method for propagating updates from main memory to the backup database on disk is also provided. The emphasis is on decoupling the I/O activities related to the propagation to disk from the forward transaction execution in memory. The authors also construct a high-level recovery manager based on operation logging on top of the page-based algorithms. The proposed algorithms are motivated by the characteristics of large MMDBs, and exploit the technology of nonvolatile RAM.

Microprocessor technology trends The rapid pace of advancement of microprocessor technology has shown no sign of diminishing, and this pace is expected to continue in the future. Recent trends in such areas as silicon technology, processor architecture and implementation, system organization, buses, higher levels of integration, self-testing, caches, coprocessors, and fault tolerance are discussed, and expectations for further ad...

Read Optimized File System Designs: A Performance Evaluation This paper presents a performance comparison of several file system allocation policies. The file systems are designed to provide high bandwidth between disks and main memory by taking advantage of parallelism in an underlying disk array, catering to large units of transfer, and minimizing the bandwidth dedicated to the transfer of meta data. All of the file systems described use a mul- tiblock allocation strategy which allows both large and small files to be allocated efficiently. Simulation results show that these multiblock policies result in systems that are able to utilize a large percentage of the underlying disk bandwidth; more than 90% in sequential cases. As general purpose systems are called upon to support more data intensive applications such as databases and super- computing, these policies offer an opportunity to provide superior performance to a larger class of users.

Storage Technology: RAID and Beyond

Object Placement in Parallel Object-Oriented Database Systems Parallelism is a viable solution to constructing high performance object-oriented database systems. In parallel systems based on a shared-nothing architecture, the database is horizontally declustered across multiple processors, enabling the system to employ multiple processors to speedup the execution time of a query. The placement of objects across the processors has a significant impact on the performance of queries that traverse a few objects. The paper describes and evaluates a greedy algorithm for the placement of objects across the processors of a system. Moreover, it describes two alternative availability strategies and quantifies their performance tradeoff using a trace-driven simulation study

Failure correction techniques for large disk arrays The ever increasing need for I/O bandwidth will be met with ever larger arrays of disks. These arrays require redundancy to protect against data loss. This paper examines alternative choices for encodings, or codes, that reliably store information in disk arrays. Codes are selected to maximize mean time to data loss or minimize disks containing redundant data, but are all constrained to minimize performance penalties associated with updating information or recovering from catastrophic disk failures. We also codes that give highly reliable data storage with low redundant data overhead for arrays of 1000 information disks.

RAID5 performance with distributed sparing Distributed sparing is a method to improve the performance of RAID5 disk arrays with respect to a dedicated sparing system with N + 2 disks (including the spare disk), since it utilizes the bandwidth of all N + 2 disks. We analyze the performance of RAID5 with distributed sparing in normal mode, degraded mode, and rebuild mode in an OLTP environment, which implies small reads and writes. The analysis in normal mode uses an M/G/1 queuing model, which takes into account the components of disk service time. In degraded mode, a low-cost approximate method is developed to estimate the mean response time of fork-join requests resulting from accesses to recreate lost data on the failed disk. Rebuild mode performance is analyzed by considering an M/G/1 vacationing server model with multiple vacations of different types to take into account differences in processing requirements for reading the first and subsequent tracks. An iterative solution method is used to estimate the mean response time of disk requests, as well as the time to read each disk, which is shown to be quite accurate through validation against simulation results. We next compare RAID5 performance in a system 1) without a cache; 2) with a cache; and 3) with a nonvolatile storage (NVS) cache. The last configuration, in addition to improved read response time due to cache hits, provides a fast-write capability, such that dirty blocks can be destaged asynchronously and at a lower priority than read requests, resulting in an improvement in read response time. The small write penalty is also reduced due to the possibility of repeated writes to dirty blocks in the cache and by taking advantage of disk geometry to efficiently destage multiple blocks at a time.

IRON file systems Commodity file systems trust disks to either work or fail completely, yet modern disks exhibit more complex failure modes. We suggest a new fail-partial failure model for disks, which incorporates realistic localized faults such as latent sector errors and block corruption. We then develop and apply a novel failure-policy fingerprinting framework, to investigate how commodity file systems react to a range of more realistic disk failures. We classify their failure policies in a new taxonomy that measures their Internal RObustNess (IRON), which includes both failure detection and recovery techniques. We show that commodity file system failure policies are often inconsistent, sometimes buggy, and generally inadequate in their ability to recover from partial disk failures. Finally, we design, implement, and evaluate a prototype IRON file system, Linux ixt3, showing that techniques such as in-disk checksumming, replication, and parity greatly enhance file system robustness while incurring minimal time and space overheads.

SafetyNet: improving the availability of shared memory multiprocessors with global checkpoint/recovery We develop an availability solution, called SafetyNet, that uses a unified, lightweight checkpoint/recovery mechanism to support multiple long-latency fault detection schemes. At an abstract level, SafetyNet logically maintains multiple, globally consistent checkpoints of the state of a shared memory multiprocessor (i.e., processors, memory, and coherence permissions), and it recovers to a pre-fault checkpoint of the system and re-executes if a fault is detected. SafetyNet efficiently coordinates checkpoints across the system in logical time and uses "logically atomic" coherence transactions to free checkpoints of transient coherence state. SafetyNet minimizes performance overhead by pipelining checkpoint validation with subsequent parallel execution.We illustrate SafetyNet avoiding system crashes due to either dropped coherence messages or the loss of an interconnection network switch (and its buffered messages). Using full-system simulation of a 16-way multiprocessor running commercial workloads, we find that SafetyNet (a) adds statistically insignificant runtime overhead in the common-case of fault-free execution, and (b) avoids a crash when tolerated faults occur.

Learning internal representations Probably the most important problem in machinelearning is the preliminary biasing of alearner's hypothesis space so that it is smallenough to ensure good generalisation fromreasonable training sets, yet large enough thatit contains a good solution to the problem beinglearnt. In this paper a mechanism for automatically learning or biasing the learner's hypothesisspace is introduced. It works by firstlearning an appropriate internal representation for a learning environment and then...

Automatic recovery from runtime failures We present a technique to make applications resilient to failures. This technique is intended to maintain a faulty application functional in the field while the developers work on permanent and radical fixes. We target field failures in applications built on reusable components. In particular, the technique exploits the intrinsic redundancy of those components by identifying workarounds consisting of alternative uses of the faulty components that avoid the failure. The technique is currently implemented for Java applications but makes little or no assumptions about the nature of the application, and works without interrupting the execution flow of the application and without restarting its components. We demonstrate and evaluate this technique on four mid-size applications and two popular libraries of reusable components affected by real and seeded faults. In these cases the technique is effective, maintaining the application fully functional with between 19% and 48% of the failure-causing faults, depending on the application. The experiments also show that the technique incurs an acceptable runtime overhead in all cases.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Target Detection in Colorful Imaging Sonar Based on HOG TKIS-I helmet-mounted colorful imaging sonar is mounted on the helmet of divers to serve as their underwater eyes. Currently, there are more than two dozens of it serving the navy of China. However, under the complex underwater environment, divers usually take great risks while performing underwater operations. The work of this paper aims to achieve automatic underwater target detection of the imaging sonar in order for divers not to dive into water. The paper combined Histogram of Oriented Gradient (HOG) in computer vision for feature extraction and support vector machine SVM) for classification to achieve quick underwater target detection. The results showed high detection rate. All these work are the foundation for automatic underwater target detection and recognition in the future.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Extracting key-substring-group features for text classification In many text classification applications, it is appealing to take every document as a string of characters rather than a bag of words. Previous research studies in this area mostly focused on different variants of generative Markov chain models. Although discriminative machine learning methods like Support Vector Machine (SVM) have been quite successful in text classification with word features, it is neither effective nor efficient to apply them straightforwardly taking all substrings in the corpus as features. In this paper, we propose to partition all substrings into statistical equivalence groups, and then pick those groups which are important (in the statistical sense) as features (named key-substring-group features) for text classification. In particular, we propose a suffix tree based algorithm that can extract such features in linear time (with respect to the total number of characters in the corpus). Our experiments on English, Chinese and Greek datasets show that SVM with key-substring-group features can achieve outstanding performance for various text classification tasks.

A Semi-Supervised Bayesian Network Model for Microblog Topic Classification. Microblogging services have brought users to a new era of knowledge dissemination and information seeking. However, the large volume and multi-aspect of messages hinder the ability of users to conveniently locate the specific messages that they are interested in. While many researchers wish to employ traditional text classification approaches to effectively understand messages on microblogging services, the limited length of the messages prevents these approaches from being employed to their full potential. To tackle this problem, we propose a novel semi-supervised learning scheme to seamlessly integrate the external web resources to compensate for the limited message length. Our approach first trains a classifier based on the available labeled data as well as some auxiliary cues mined from the web, and probabilistically predicts the categories for all unlabeled data. It then trains a new classifier using the labels for all messages and the auxiliary cues, and iterates the process to convergence. Our approach not only greatly reduces the time-consuming and labor-intensive labeling process, but also deeply exploits the hidden information from unlabeled data and related text resources. We conducted extensive experiments on two real-world microblogging datasets. The results demonstrate the effectiveness of the proposed approaches which produce promising performance as compared to state-of-the-art methods. © 2012 The COLING.

The bag-of-repeats representation of documents n-gram representations of documents may improve over a simple bag-of-word representation by relaxing the independence assumption of word and introducing context. However, this comes at a cost of adding features which are non-descriptive, and increasing the dimension of the vector space model exponentially. We present new representations that avoid both pitfalls. They are based on sound theoretical notions of stringology, and can be computed in optimal asymptotic time with algorithms using data structures from the suffix family. While maximal repeats have been used in the past for similar tasks, we show how another equivalence class of repeats -- largest-maximal repeats -- obtain similar or better results, with only a fraction of the features. This class acts as a minimal generative basis of all repeated substrings. We also report their use for topic modeling, showing easier to interpret models.

Robust Graph Mode Seeking by Graph Shift

Using decision tree for diagnosing heart disease patients Heart disease is the leading cause of death in the world over the past 10 years. Researchers have been using several data mining techniques to help health care professionals in the diagnosis of heart disease. Decision Tree is one of the successful data mining techniques used. However, most research has applied J4.8 Decision Tree, based on Gain Ratio and binary discretization. Gini Index and Information Gain are two other successful types of Decision Trees that are less used in the diagnosis of heart disease. Also other discretization techniques, voting method, and reduced error pruning are known to produce more accurate Decision Trees. This research investigates applying a range of techniques to different types of Decision Trees seeking better performance in heart disease diagnosis. A widely used benchmark data set is used in this research. To evaluate the performance of the alternative Decision Trees the sensitivity, specificity, and accuracy are calculated. The research proposes a model that outperforms J4.8 Decision Tree and Bagging algorithm in the diagnosis of heart disease patients.

Studies of the onset and persistence of medical concerns in search logs The Web provides a wealth of information about medical symptoms and disorders. Although this content is often valuable to consumers, studies have found that interaction with Web content may heighten anxiety and stimulate healthcare utilization. We present a longitudinal log-based study of medical search and browsing behavior on the Web. We characterize how users focus on particular medical concerns and how concerns persist and influence future behavior, including changes in focus of attention in searching and browsing for health information. We build and evaluate models that predict transitions from searches on symptoms to searches on health conditions, and escalations from symptoms to serious illnesses. We study the influence that the prior onset of concerns may have on future behavior, including sudden shifts back to searching on the concern amidst other searches. Our findings have implications for refining Web search and retrieval to support people pursuing diagnostic information.

Predicting individual disease risk based on medical history The monumental cost of health care, especially for chronic disease treatment, is quickly becoming unmanageable. This crisis has motivated the drive towards preventative medicine, where the primary concern is recognizing disease risk and taking action at the earliest signs. However, universal testing is neither time nor cost efficient. We propose CARE, a Collaborative Assessment and Recommendation Engine, which relies only on a patient's medical history using ICD-9-CM codes in order to predict future diseases risks. CARE uses collaborative filtering to predict each patient's greatest disease risks based on their own medical history and that of similar patients. We also describe an Iterative version, ICARE, which incorporates ensemble concepts for improved performance. These novel systems require no specialized information and provide predictions for medical conditions of all kinds in a single run. We present experimental results on a Medicare dataset, demonstrating that CARE and ICARE perform well at capturing future disease risks.

The Curse of Highly Variable Functions for Local Kernel Machines We present a series of theoretical arguments supporting the claim that a large class of modern learning algorithms that rely solely on the smooth- ness prior - with similarity between examples expressed with a local kernel - are sensitive to the curse of dimensionality, or more precisely to the variability of the target. Our discussion covers supervised, semi- supervised and unsupervised learning algorithms. These algorithms are found to be local in the sense that crucial properties of the learned func- tion at x depend mostly on the neighbors of x in the training set. This makes them sensitive to the curse of dimensionality, well studied for classical non-parametric statistical learning. We show in the case of the Gaussian kernel that when the function to be learned has many variations, these algorithms require a number of training examples proportional to the number of variations, which could be large even though there may ex- ist short descriptions of the target function, i.e. their Kolmogorov com- plexity may be low. This suggests that there exist non-local learning algorithms that at least have the potential to learn about such structured but apparently complex functions (because locally they have many vari- ations), while not using very specific prior domain knowledge.

Links between perceptrons, MLPs and SVMs We propose to study links between three important classification algorithms: Perceptrons, Multi-Layer Perceptrons (MLPs) and Support Vector Machines (SVMs). We first study ways to control the capacity of Perceptrons (mainly regularization parameters and early stopping), using the margin idea introduced with SVMs. After showing that under simple conditions a Perceptron is equivalent to an SVM, we show it can be computationally expensive in time to train an SVM (and thus a Perceptron) with stochastic gradient descent, mainly because of the margin maximization term in the cost function. We then show that if we remove this margin maximization term, the learning rate or the use of early stopping can still control the margin. These ideas are extended afterward to the case of MLPs. Moreover, under some assumptions it also appears that MLPs are a kind of mixture of SVMs, maximizing the margin in the hidden layer space. Finally, we present a very simple MLP based on the previous findings, which yields better performances in generalization and speed than the other models.

Regularized Auto-Encoders Estimate Local Statistics

Complexity of Power Default Reasoning This paper derives a new and surprisingly low complexity result for inference in a new form of Reiter's propositional default logic. The problem studied here is the "default inference problem" whose fundamental importance was pointed out by Kraus, Lehmann, and Magidor. We prove that ``normal'' default inference, in propositional logic, is a problem complete for co-NP(3), the third level of the so-called Boolean hierarchy. Our result (by changing the underlying semantics) contrasts favorably with a similar result of Gottlob, who proves that standard default inference is complete for the second level of the polynomial hierarchy. Our inference relation also obeys all of the laws for preferential consequence relations set forth by Kraus, Lehmann, and Magidor. In particular, we get the property of being able to reason by cases and the law of cautious monotony. Both of these laws fail for standard propositional default logic.The key technique for our results is the use of Scott's domain theory to integrate defaults into partial model theory of the logic, instead of keeping defaults as quasi-proof rules in the syntax. In particular, reasoning disjunctively entails using the Smyth powerdomain.

Meta-ViPIOS: Harness Distributed I/O Resources with ViPIOS Two factors strongly inuenced the research in high performancecomputing in the last few years, the I/O bottleneckand cluster systems. Firstly, for many supercomputing applicationsthe limiting factor is not the number of availableCPUs anymore, but the bandwidth of the disk I/O system.Secondly, a shift from the classical, costly supercomputersystems to aordable clusters of workstations is apparent,which allows problem solutions to a much lower price.As a result we present in this paper...

P-Selectivity, immunity, and the power of one bit We prove that P-sel, the class of all P-selective sets, is EXP-immune, but is not EXP/1-immune. That is, we prove that some infinite P-selective set has no infinite EXP-time subset, but we also prove that every infinite P-selective set has some infinite subset in EXP/1. Informally put, the immunity of P-sel is so fragile that it is pierced by a single bit of information. The above claims follow from broader results that we obtain about the immunity of the P-selective sets. In particular, we prove that for every recursive function f, P-sel is DTIME(f)-immune. Yet we also prove that P-sel is not ${\it \Pi}^{p}_{2}$/1-immune.

"The sum of all human knowledge": A systematic review of scholarly research on the content of Wikipedia AbstractWikipedia may be the best-developed attempt thus far to gather all human knowledge in one place. Its accomplishments in this regard have made it a point of inquiry for researchers from different fields of knowledge. A decade of research has thrown light on many aspects of the Wikipedia community, its processes, and its content. However, due to the variety of fields inquiring about Wikipedia and the limited synthesis of the extensive research, there is little consensus on many aspects of Wikipedia's content as an encyclopedic collection of human knowledge. This study addresses the issue by systematically reviewing 110 peer-reviewed publications on Wikipedia content, summarizing the current findings, and highlighting the major research trends. Two major streams of research are identified: the quality of Wikipedia content including comprehensiveness, currency, readability, and reliability and the size of Wikipedia. Moreover, we present the key research trends in terms of the domains of inquiry, research design, data source, and data gathering methods. This review synthesizes scholarly understanding of Wikipedia content and paves the way for future studies.

Learning Recurrent Binary/Ternary Weights. Recurrent neural networks (RNNs) have shown excellent performance in processing sequence data. However, they are both complex and memory intensive due to their recursive nature. These limitations make RNNs difficult to embed on mobile devices requiring real-time processes with limited hardware resources. To address the above issues, we introduce a method that can learn binary and ternary weights during the training phase to facilitate hardware implementations of RNNs. As a result, using this approach replaces all multiply-accumulate operations by simple accumulations, bringing significant benefits to custom hardware in terms of silicon area and power consumption. On the software side, we evaluate the performance (in terms of accuracy) of our method using long short-term memories (LSTMs) on various sequential models including sequence classification and language modeling. We demonstrate that our method achieves competitive results on the aforementioned tasks while using binary/ternary weights during the runtime. On the hardware side, we present custom hardware for accelerating the recurrent computations of LSTMs with binary/ternary weights. Ultimately, we show that LSTMs with binary/ternary weights can achieve up to 12x memory saving and 10x inference speedup compared to the full-precision implementation on an ASIC platform.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Feature learning and deep architectures: new directions for music informatics As we look to advance the state of the art in content-based music informatics, there is a general sense that progress is decelerating throughout the field. On closer inspection, performance trajectories across several applications reveal that this is indeed the case, raising some difficult questions for the discipline: why are we slowing down, and what can we do about it? Here, we strive to address both of these concerns. First, we critically review the standard approach to music signal analysis and identify three specific deficiencies to current methods: hand-crafted feature design is sub-optimal and unsustainable, the power of shallow architectures is fundamentally limited, and short-time analysis cannot encode musically meaningful structure. Acknowledging breakthroughs in other perceptual AI domains, we offer that deep learning holds the potential to overcome each of these obstacles. Through conceptual arguments for feature learning and deeper processing architectures, we demonstrate how deep processing models are more powerful extensions of current methods, and why now is the time for this paradigm shift. Finally, we conclude with a discussion of current challenges and the potential impact to further motivate an exploration of this promising research area.

Audio Chord Recognition with Recurrent Neural Networks.

Learning Semantic Representations for the Phrase Translation Model. This paper presents a novel semantic-based phrase translation model. A pair of source and target phrases are projected into continuous-valued vector representations in a low-dimensional latent semantic space, where their translation score is computed by the distance between the pair in this new space. The projection is performed by a multi-layer neural network whose weights are learned on parallel training data. The learning is aimed to directly optimize the quality of end-to-end machine translation results. Experimental evaluation has been performed on two Europarl translation tasks, English-French and German-English. The results show that the new semantic-based phrase translation model significantly improves the performance of a state-of-the-art phrase-based statistical machine translation sys-tem, leading to a gain of 0.7-1.0 BLEU points.

Convex Two-Layer Modeling. Latent variable prediction models, such as multi-layer networks, impose auxiliary latent variables between inputs and outputs to allow automatic inference of implicit features useful for prediction. Unfortunately, such models are difficult to train because inference over latent variables must be performed concurrently with parameter optimization---creating a highly non-convex problem. Instead of proposing another local training method, we develop a convex relaxation of hidden-layer conditional models that admits global training. Our approach extends current convex modeling approaches to handle two nested nonlinearities separated by a non-trivial adaptive latent layer. The resulting methods are able to acquire two-layer models that cannot be represented by any single-layer model over the same features, while improving training quality over local heuristics.

Moving Beyond Feature Design: Deep Architectures and Automatic Feature Learning in Music Informatics.

Learning Continuous Phrase Representations For Translation Modeling This paper tackles the sparsity problem in estimating phrase translation probabilities by learning continuous phrase representations, whose distributed nature enables the sharing of related phrases in their representations. A pair of source and target phrases are projected into continuous-valued vector representations in a low-dimensional latent space, where their translation score is computed by the distance between the pair in this new space. The projection is performed by a neural network whose weights are learned on parallel training data. Experimental evaluation has been performed on two WMT translation tasks. Our best result improves the performance of a state-of-the-art phrase-based statistical machine translation system trained on WMT 2012 French-English data by up to 1.3 BLEU points.

A Scalable Hierarchical Distributed Language Model Neural probabilistic language models (NPLMs) have been shown to be competi- tive with and occasionally superior to the widely-used n-gram language models. The main drawback of NPLMs is their extremely long training and testing times. Morin and Bengio have proposed a hierarchical language model built around a binary tree of words, which was two orders of magnitude faster than the non- hierarchical model it was based on. However, it performed considerably worse than its non-hierarchical counterpart in spite of using a wo rd tree created using expert knowledge. We introduce a fast hierarchical language model along with a simple feature-based algorithm for automatic construction of word trees from the data. We then show that the resulting models can outperform non-hierarchical neural models as well as the best n-gram models.

Local deep feature learning framework for 3D shape. For 3D shape analysis, an effective and efficient feature is the key to popularize its applications in 3D domain. In this paper, we present a novel framework to learn and extract local deep feature (LDF), which encodes multiple low-level descriptors and provides high-discriminative representation of local region on 3D shape. The framework consists of four main steps. First, several basic descriptors are calculated and encapsulated to generate geometric bag-of-words in order to make full use of the various basic descriptors׳ properties. Then 3D mesh is down-sampled to hundreds of feature points for accelerating the model learning. Next, in order to preserve the local geometric information and establish the relationships among points in a local area, the geometric bag-of-words are encoded into local geodesic-aware bag-of-features (LGA-BoF). However, the resulting feature is redundant, which leads to low discriminative and efficiency. Therefore, in the final step, we use deep belief networks (DBNs) to learn a model, and use it to generate the LDF, which is high-discriminative and effective for 3D shape applications. 3D shape correspondence and symmetry detection experiments compared with related feature descriptors are carried out on several datasets and shape recognition is also conducted, validating the proposed local deep feature learning framework.

The Diffculty of Training Deep Architectures and the Effect of Unsupervised Pre-Training Whereas theoretical work suggests that deep ar- chitectures might be more e cient at represent- ing highly-varying functions, training deep ar- chitectures was unsuccessful until the recent ad- vent of algorithms based on unsupervised pre- training. Even though these new algorithms have enabled training deep models, many questions remain as to the nature of this di cult learning problem. ...

Best Practices for Convolutional Neural Networks Applied to Visual Document Analysis Neural networks are a powerful technology forclassification of visual inputs arising from documents.However, there is a confusing plethora of different neuralnetwork methods that are used in the literature and inindustry. This paper describes a set of concrete bestpractices that document analysis researchers can use toget good results with neural networks. The mostimportant practice is getting a training set as large aspossible: we expand the training set by adding a newform of distorted data. The next most important practiceis that convolutional neural networks are better suited forvisual document tasks than fully connected networks. Wepropose that a simple "do-it-yourself" implementation ofconvolution with a flexible architecture is suitable formany visual document problems. This simpleconvolutional neural network does not require complexmethods, such as momentum, weight decay, structure-dependentlearning rates, averaging layers, tangent prop,or even finely-tuning the architecture. The end result is avery simple yet general architecture which can yieldstate-of-the-art performance for document analysis. Weillustrate our claims on the MNIST set of English digitimages.

Parallel database systems: the future of high performance database systems

Consistency without ordering Modern file systems use ordering points to maintain consistency in the face of system crashes. However, such ordering leads to lower performance, higher complexity, and a strong and perhaps naive dependence on lower layers to correctly enforce the ordering of writes. In this paper, we introduce the No-Order File System (NoFS), a simple, lightweight file system that employs a novel technique called backpointer-based consistency to provide crash consistency without ordering writes as they go to disk. We utilize a formal model to prove that NoFS provides data consistency in the event of system crashes; we show through experiments that NoFS is robust to such crashes, and delivers excellent performance across a range of workloads. Backpointer-based consistency thus allows NoFS to provide crash consistency without resorting to the heavyweight machinery of traditional approaches.

On Approaches to Explaining Infeasibility of Sets of Boolean Clauses These last years, the issue of locating and explaining contradictions inside sets of propositional clauses has received a renewed attention due to the emergence of very efficient SAT solvers. In case of inconsistency, many such solvers merely conclude that no solution exists or provide an upper approximation of the subset of clauses that are contradictory. However, in most application domains, only knowing that a problem does not admit any solution is not enough informative, and it is important to know which clauses are actually conflicting. In this paper, the focus is on the concept of minimally unsatisfiable subformulas (MUSes), which explain logical inconsistency in terms of minimal sets of contradictory clauses. Specifically, various recent results and computational approaches about MUSes and related concepts are discussed.

Boosting MUC extraction in unsatisfiable constraint networks One very fertile domain of applied Artificial Intelligence is constraint solving technologies. Especially, constraint networks that concern problems that can be represented using discrete variables, together with constraints on allowed instantiation values for these variables. Every solution to a constraint network must satisfy every constraint. When no solution exists, the user might want to know the actual reasons leading to the absence of global solution. In this respect, extracting mucs (Minimal Unsatisfiable Cores) from an unsatisfiable constraint network is a useful process when causes of unsatisfiability must be understood so that the network can be re-engineered and relaxed to become satisfiable. Despite bad worst-case computational complexity results, various muc-finding approaches that appear tractable for many real-life instances have been proposed. Many of them are based on the successive identification of so-called transition constraints. In this respect, we show how local search can be used to possibly extract additional transition constraints at each main iteration step. In the general constraint networks setting, the approach is shown to outperform a technique based on a form of model rotation imported from the sat-related technology and that also exhibits additional transition constraints. Our extensive computational experimentations show that this enhancement also boosts the performance of state-of-the-art DC(WCORE)-like MUC extractors.

Scheduler-Assisted Prefetching: Efficient Demand Paging for Embedded Systems Embedded systems tend to use demand paging in order to provide more memory to applications in a cost-effective manner. However, demand paging drastically degrades the performance when the page fault rate is high. Prefetching has been known as a common remedy for page fault overhead. Although many prefetching mechanisms have been proposed, they are either effective only for specific page access patterns or too straight-forward to decrease a page fault rate to an acceptable level. We propose a scheduler-assisted prefetching mechanism which does not have such fundamental defects. As a proof of concept, our mechanism was completely implemented in Linux. We have also conducted a series of experiments to show its effectiveness. The experimental results showed a significant improvement: the number of the major page faults and the scheduling latency decreased by 30% and 51%, respectively.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

WenZher: comprehensive vertical search for healthcare domain Online health seeking has transformed the way of health knowledge exchange and reusability. The existing general and vertical health search engines, however, just routinely return lists of matched documents or question answer (QA) pairs, which may overwhelm the seekers or not sufficiently meet the seekers' expectations. Instead, our multilingual system is able to return one multi-faceted answer that is well-structured and precisely extracted from multiple heterogeneous healthcare sources. Further, should the seekers not be satisfied with the returned search results, our system can automatically route the unsolved questions to the professionals with relevant expertise.

Modeling Disease Progression via Fused Sparse Group Lasso. Alzheimer's Disease (AD) is the most common neurodegenerative disorder associated with aging. Understanding how the disease progresses and identifying related pathological biomarkers for the progression is of primary importance in the clinical diagnosis and prognosis of Alzheimer's disease. In this paper, we develop novel multi-task learning techniques to predict the disease progression measured by cognitive scores and select biomarkers predictive of the progression. In multi-task learning, the prediction of cognitive scores at each time point is considered as a task, and multiple prediction tasks at different time points are performed simultaneously to capture the temporal smoothness of the prediction models across different time points. Specifically, we propose a novel convex fused sparse group Lasso (cFSGL) formulation that allows the simultaneous selection of a common set of biomarkers for multiple time points and specific sets of biomarkers for different time points using the sparse group Lasso penalty and in the meantime incorporates the temporal smoothness using the fused Lasso penalty. The proposed formulation is challenging to solve due to the use of several non-smooth penalties. One of the main technical contributions of this paper is to show that the proximal operator associated with the proposed formulation exhibits a certain decomposition property and can be computed efficiently; thus cFSGL can be solved efficiently using the accelerated gradient method. To further improve the model, we propose two non-convex formulations to reduce the shrinkage bias inherent in the convex formulation. We employ the difference of convex (DC) programming technique to solve the non-convex formulations. We have performed extensive experiments using data from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Results demonstrate the effectiveness of the proposed progression models in comparison with existing methods for disease progression. We also perform longitudinal stability selection to identify and analyze the temporal patterns of biomarkers in disease progression.

Personalized Recommendations of Locally Interesting Venues to Tourists via Cross-Region Community Matching You are in a new city. You are not familiar with the places and neighborhoods. You want to know all about the exciting sights, food outlets, and cultural venues that the locals frequent, in particular those that suit your personal interests. Even though there exist many mapping, local search, and travel assistance sites, they mostly provide popular and famous listings such as Statue of Liberty and Eiffel Tower, which are well-known places but may not suit your personal needs or interests. Therefore, there is a gap between what tourists want and what dominant tourism resources are providing. In this work, we seek to provide a solution to bridge this gap by exploiting the rich user-generated location contents in location-based social networks in order to offer tourists the most relevant and personalized local venue recommendations. In particular, we first propose a novel Bayesian approach to extract the social dimensions of people at different geographical regions to capture their latent local interests. We next mine the local interest communities in each geographical region. We then represent each local community using aggregated behaviors of community members. Finally, we correlate communities across different regions and generate venue recommendations to tourists via cross-region community matching. We have sampled a representative subset of check-ins from Foursquare and experimentally verified the effectiveness of our proposed approaches.

The bag-of-repeats representation of documents n-gram representations of documents may improve over a simple bag-of-word representation by relaxing the independence assumption of word and introducing context. However, this comes at a cost of adding features which are non-descriptive, and increasing the dimension of the vector space model exponentially. We present new representations that avoid both pitfalls. They are based on sound theoretical notions of stringology, and can be computed in optimal asymptotic time with algorithms using data structures from the suffix family. While maximal repeats have been used in the past for similar tasks, we show how another equivalence class of repeats -- largest-maximal repeats -- obtain similar or better results, with only a fraction of the features. This class acts as a minimal generative basis of all repeated substrings. We also report their use for topic modeling, showing easier to interpret models.

Extracting key-substring-group features for text classification In many text classification applications, it is appealing to take every document as a string of characters rather than a bag of words. Previous research studies in this area mostly focused on different variants of generative Markov chain models. Although discriminative machine learning methods like Support Vector Machine (SVM) have been quite successful in text classification with word features, it is neither effective nor efficient to apply them straightforwardly taking all substrings in the corpus as features. In this paper, we propose to partition all substrings into statistical equivalence groups, and then pick those groups which are important (in the statistical sense) as features (named key-substring-group features) for text classification. In particular, we propose a suffix tree based algorithm that can extract such features in linear time (with respect to the total number of characters in the corpus). Our experiments on English, Chinese and Greek datasets show that SVM with key-substring-group features can achieve outstanding performance for various text classification tasks.

A multi-task learning formulation for predicting disease progression Alzheimer's Disease (AD), the most common type of dementia, is a severe neurodegenerative disorder. Identifying markers that can track the progress of the disease has recently received increasing attentions in AD research. A definitive diagnosis of AD requires autopsy confirmation, thus many clinical/cognitive measures including Mini Mental State Examination (MMSE) and Alzheimer's Disease Assessment Scale cognitive subscale (ADAS-Cog) have been designed to evaluate the cognitive status of the patients and used as important criteria for clinical diagnosis of probable AD. In this paper, we propose a multi-task learning formulation for predicting the disease progression measured by the cognitive scores and selecting markers predictive of the progression. Specifically, we formulate the prediction problem as a multi-task regression problem by considering the prediction at each time point as a task. We capture the intrinsic relatedness among different tasks by a temporal group Lasso regularizer. The regularizer consists of two components including an L2,1-norm penalty on the regression weight vectors, which ensures that a small subset of features will be selected for the regression models at all time points, and a temporal smoothness term which ensures a small deviation between two regression models at successive time points. We have performed extensive evaluations using various types of data at the baseline from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database for predicting the future MMSE and ADAS-Cog scores. Our experimental studies demonstrate the effectiveness of the proposed algorithm for capturing the progression trend and the cross-sectional group differences of AD severity. Results also show that most markers selected by the proposed algorithm are consistent with findings from existing cross-sectional studies.

Multimedia answering: enriching text QA with media information Existing community question-answering forums usually provide only textual answers. However, for many questions, pure texts cannot provide intuitive information, while image or video contents are more appropriate. In this paper, we introduce a scheme that is able to enrich text answers with image and video information. Our scheme investigates a rich set of techniques including question/answer classification, query generation, image and video search reranking, etc. Given a question and the community-contributed answer, our approach is able to determine which type of media information should be added, and then automatically collects data from Internet to enrich the textual answer. Different from some efforts that attempt to directly answer questions with image and video data, our approach is built based on the community-contributed textual answers and thus it is more feasible and able to deal with more complex questions. We have conducted empirical study on more than 3,000 QA pairs and the results demonstrate the effectiveness of our approach.

Learning Compact Hash Codes for Multimodal Representations using Orthogonal Deep Structure As large-scale multimodal data are ubiquitous in many real-world applications, learning multimodal representations for efficient retrieval is a fundamental problem. Most existing methods adopt shallow structures to perform multimodal representation learning. Due to a limitation of learning ability of shallow structures, they fail to capture the correlation of multiple modalities. Recently, multimodal deep learning was proposed and had proven its superiority in representing multimodal data due to its high nonlinearity. However, in order to learn compact and accurate representations, how to reduce the redundant information lying in the multimodal representations and incorporate different complexities of different modalities in the deep models is still an open problem. In order to address the aforementioned problem, we propose a hashing-based orthogonal deep model to learn accurate and compact multimodal representations in this paper. The method can better capture the intra-modality and inter-modality correlations to learn accurate representations. Meanwhile, in order to make the representations compact, the hashing-based model can generate compact hash codes and the proposed orthogonal structure can reduce the redundant information lying in the codes by imposing orthogonal regularizer on the weighting matrices. We also theoretically prove that in this case the learned codes are guaranteed to be approximately orthogonal. Moreover, considering the different characteristics of different modalities, effective representations can be attained with different number of layers for different modalities. Comprehensive experiments on three real-world datasets demonstrate a substantial gain of our method on retrieval tasks compared with existing algorithms.

LIBSVM: A library for support vector machines LIBSVM is a library for Support Vector Machines (SVMs). We have been actively developing this package since the year 2000. The goal is to help users to easily apply SVM to their applications. LIBSVM has gained wide popularity in machine learning and many other areas. In this article, we present all implementation details of LIBSVM. Issues such as solving SVM optimization problems theoretical convergence multiclass classification probability estimates and parameter selection are discussed in detail.

A Sparse and Locally Shift Invariant Feature Extractor Applied to Document Images We describe an unsupervised learning algorithm for ex- tracting sparse and locally shift-invariant features. We also devise a principled procedure for learning hierarchies of in- variant features. Each feature detector is composed of a set of trainable convolutional filters followed by a max-pooling layer over non-overlapping windows, and a point-wise sig- moid non-linearity. A second stage of more invariant fea- tures is fed with patches provided by the first stage feature extractor, and is trained in the same way. The method is used to pre-train the first four layers of a deep convolutional network which achieves state-of-the-art performance on the MNIST dataset of handwritten digits. The final testing error rate is equal to 0.42%. Preliminary experiments on com- pression of bitonal document images show very promising results in terms of compression ratio and reconstruction er- ror.

Weighted voting for replicated data In a new algorithm for maintaining replicated data, every copy of a replicated file is assigned some number of votes. Every transaction collects a read quorum of rvotes to read a file, and a write quorum of wvotes to write a file, such that r+w is greater than the total number of votes assigned to the file. This ensures that there is a non-null intersection between every read quorum and every write quorum. Version numbers make it possible to determine which copies are current. The reliability and performance characteristics of a replicated file can be controlled by appropriately choosing r, w, and the file's voting configuration. The algorithm guarantees serial consistency, admits temporary copies in a natural way by the introduction of copies with no votes, and has been implemented in the context of an application system called Violet.

Boosting a complete technique to find MSS and MUS thanks to a local search oracle In this paper, a new complete technique to compute Maximal Satisfiable Subsets (MSS) and Minimally Unsatisfiable Subformulas (MUS) of sets of Boolean clauses is introduced. The approach improves the currently most efficient complete technique in several ways. It makes use of the powerful concept of critical clause and of a computationally inexpensive local search oracle to boost an exhaustive algorithm proposed by Liffiton and Sakallah. These features can allow exponential efficiency gains to be obtained. Accordingly, experimental studies show that this new approach outperforms the best current existing exhaustive ones.

Hierarchical Representation Using NMF.

Optimizing large data transfers in parity-declustered data layouts. Disk arrays allow faster access to users' data by distributing the data among a collection of disks and allowing parallel access. Fault tolerance in a disk array can be achieved by using a data layout, and the technique of parity declustering allows faster failure recovery at the cost of additional space dedicated to redundant information. A collection of six performance conditions that parity-declustered data layouts should satisfy has guided most previous work; however two of these conditions (Maximal parallelism and Large write optimization) cannot be jointly satisfied in most cases. This limits the ability of parity-declustered data layouts to take full advantage of the available parallelism during large data transfers. We present data layouts that approximately satisfy these two conditions simultaneously for all possible array configurations, and bound the deviations from complete satisfaction. Our results yield improved performance guarantees for large data transfers in parity-declustered data layouts.

Connections between the complexity of unique satisfiability and the threshold behavior of randomized reductions The present research is motivated by new results on the complexity of the unique satisfiability problem (USAT). Some new results are obtained, using the concept of randomized reductions. The proofs use only the fact that USAT is complete for DP under randomized reductions, even though the probability bound of these reductions may be low. Furthermore, the results show that the structural complexities of USAT and DP many-one complete sets are very similar, lending support to the argument that even sets complete under `weak' randomized reductions can capture the properties of the many-one complete sets. The authors generalize these results for the Boolean hierarchy and give upper and lower bounds on the thresholds for these classes

On the structure of bounded queries to arbitrary NP sets In [Kad87b], Kadin showed that if the Polynomial Hierarchy (PH) has infinitely many levels, then for all $k$, $P^{SAT[k]} \subseteq P^{SAT[k+1]}$. In this paper, we extend Kadin''s technique to show that a proper query hierarchy is not an exclusive property of SAT. In fact, for any $A \in NP \overbrace{low_{3}}$, if PH is infinite, then $P^{A[k]} \subseteq P^{A[k+1]}$. Moreover, for the case of parallel queries, we show that $P^{A||[k+1]}$ is not contained in $P^{SAT||[k]}$. We claim that having a proper query hierarchy is a consequence of the oracle access mechanism and not a result of the ``hardness'''' of a set. To support this claim, we show that assuming PH is infinite, one can construct an intermediate set $B \in NP$ so that $P^{B[k+1]} \subseteq P^{SAT[k]}$. That is, the query hierarchy for $B$ grows as ``tall'''' as the query hierarchy for SAT. In addition, $B$ is intermediate, so it is not ``hard'''' in any sense (e.g., not NP hard under many-one, Turing, or strong nondeterministic reductions). Using these same techniques, we explore some other questions about query hierarchies. For example, we show that is there exists any $A$ such that $P^{A[2]} = P^{SAT[1]}$ then PH collapses to $\Delta^{P}_{3}$.

On the Power of Deterministic Reductions to C=P The counting class C = P, which captures the notion of "exact counting," while extremely powerful under various nondeterministic reductions, is quite weak under polynomial-time deterministic reductions. We discuss the analogies between NP and co-C = P, which allow us to derive many interesting results for such deterministic reductions to co-C = P. We exploit these results to obtain some interesting oracle separations. Most importantly, we show that there exists an oracle A such that +P(A) not-subset-or-equal-to P(C=PA) and BPP(A) not-subset-or-equal-to P(C=PA). Therefore, techniques that would prove that C = P and PP are polynomial-time Turing equivalent, or that C = P is polynomial-time Turing hard for the polynomial-time hierarchy, would not relativize.

Polynomial terse sets Let A be a set and k ∈ N be such that we wish to know the answers to x 1 ∈ A ?, x 2 ∈ A ?, …, x k ∈ A ? for various k -tuples 〈 x 1 , x 2 , …, x k 〉. If this problem requires k queries to A in order to be solved in polynomial time then A is called polynomial terse or pterse . We show the existence of both arbitrarily complex pterse and non-pterse sets; and that P ≠ NP iff every NP-complete set is pterse. We also show connections with p -immunity, p -selective, p -generic sets, and the boolean hierarchy. In our framework unique satisfiability (and a variation of it called k SAT is, in some sense, “close” to satisfiability.

The complexity of combinatorial problems with succinct input representation Several languages for the succinct representation of the instances of combinatorial problems are investigated. These languages have been introduced in [20, 2] and [5] where it has been shown that describing the instances by these languages causes a blow-up of the complexities of some problems. In the present paper the descriptional power of these languages is compared by estimating the complexities of some combinatorial problems in terms of completeness in suitable classes of the “counting polynomial-time hierarchy” which is introduced here. It turns out that some of the languages are not comparable, unless P=NP Some problems left open in [2] are solved.

On the Boolean closure of NP By endowing usual nondeterministic Turing machines with new modes of acceptance we introduce new machines whose computational power is bounded by that of alternating Turing machines making only one alternation. The polynomial time classes of these machines are exactly the levels of the Boolean closure of NP which can be defined in a natural way. For all these classes natural problems can be found which are proved to be m P -complete in these classes.

Downward Collapse from a Weaker Hypothesis Hemaspaandra et al. (1997) proved that, for m i 0 and 0 i k - 1: if DIFF_s(\Sigma^p_i) \bold\Delta DIFF_m(\Sigma^p_k) is closed under complementation, then DIFF_m(\Sigma^p_k) = coDIFF_m(\Sigma^p_k).

Why not negation by fixpoint? There is a fixpoint semantics for DATALOG programs with negation that is a natural generalization of the standard semantics for DATALOG programs without negation. We show that, unfortunately, several compelling complexity-theoretic obstacles rule out its efficient implementation. As an alternative, we propose Inflationary DATALOG, an efficiently implementable cemantics for negation,based on inflationarv flxpoints.

The Complexity of Read-Once Resolution We investigate the complexity of deciding whether a propositional formula has a read-once resolution proof. We give a new and general proof of Iwama–Miynano's theorem which states that the problem whether a formula has a read-once resolution proof is iNP-complete. Moreover, we show for fixed ik&ges;2 that the additional restriction that in each resolution step one of the parent clauses is a ik-clause preserves the iNP-completeness. If we demand that the formulas are minimal unsatisfiable and read-once refutable then the problem remains iNP-complete. For the subclasses iMU(ik) of minimal unsatisfiable formulas we present a pol-time algorithm deciding whether a iMU(ik)-formula has a read-once resolution proof. Furthermore, we show that the problems whether a formula contains a iMU(ik)-subformula or a read-once refutable iMU(ik)-subformula are iNP-complete.

Affinity analysis of coded data sets Coded data sets are commonly used as compact representations of real world processes. Such data sets have been studied within various research fields from association mining, data warehousing, knowledge discovery, collaborative filtering to machine learning. However, previous studies on coded data sets have introduced methods for the analysis of rather small data sets. This study proposes applying information retrieval for enabling high performance analysis of data masses that scale beyond traditional approaches. Part of this PHD study focuses on new type of kernel projection functions that can be used to find similarities in spare discrete data spaces. This study presents experimental results how information retrieval indexes scale and outperform two common relational data schemas with a leading commercial DBMS for market basket analysis.

A performance evaluation of RAID architectures In today's computer systems, the disk I/O subsystem is often identified as the major bottleneck to system performance. One proposed solution is the so called redundant array of inexpensive disks (RAID). We examine the performance of two of the most promising RAID architectures, the mirrored array and the rotated parity array. First, we propose several scheduling policies for the mirrored array and a new data layout, group-rotate declustering, and compare their performance with each other and in combination with other data layout schemes. We observe that a policy that routes reads to the disk with the smallest number of requests provides the best performance, especially when the load on the I/O system is high. Second, through a combination of simulation and analysis, we compare the performance of this mirrored array architecture to the rotated parity array architecture. This latter study shows that: 1) given the same storage capacity (approximately double the number of disks), the mirrored array considerably outperforms the rotated parity array; and 2) given the same number of disks, the mirrored array still outperforms the rotated parity array in most cases, even for applications where I/O requests are for large amounts of data. The only exception occurs when the I/O size is very large; most of the requests are writes, and most of these writes perform full stripe write operations

Continuous retrieval of multimedia data using parallelism Most implementations of workstation-based multimedia information systems cannot support a continuous display of high resolution audio and video data and suffer from frequent disruptions and delays termed hiccups. This is due to the low I/O bandwidth of the current disk technology, the high bandwidth requirement of multimedia objects, and the large size of these objects, which requires them to be almost always disk resident. A parallel multimedia information system and the key technical ideas that enable it to support a real-time display of multimedia objects are described. In this system, a multimedia object across several disk drives is declustered, enabling the system to utilize the aggregate bandwidth of multiple disks to retrieve an object in real-time. Then, the workload of an application is distributed evenly across the disk drives to maximize the processing capability of the system. To support simultaneous display of several multimedia objects for different users, two alternative approaches are described. The first approach multitasks a disk drive among several requests while the second replicates the data and dedicates resources to each individual request. The trade-offs associated with each approach are investigated using a simulation model.

Bayesian learning and evolutionary parameter optimization In this paper I want to argue that the combination of evolutionary algorithms and neural networks can be fruitful in several ways. When estimating a functional relationship on the basis of empirical data we face three basic problems. Firstly, we have to deal with noisy and finite-sized data sets which is usually done be regularization techniques, for example Bayesian learning. Secondly, for many applications we need to encode the problem by features and have to decide which and how many of them to use. Bearing in mind the empty space phenomenon, it is often an advantage to select few features and estimate a non-linear function in a low-dimensional space. Thirdly, if we have trained several networks, we are left with the problem of model selection. These problems can be tackled by integrating several stochastic methods into an evolutionary search algorithm. The search can be designed such that it explores the parameter space to find regions corresponding to networks with a high posterior probability of being a model for the process, that generated the data. The benefits of the approach are demonstrated in detail on a regression and a classification problem. On a larger benchmark set the results are compared to other machine learning methods as Support Vector Machines.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Efficiently identifying working sets in block I/O streams Identifying groups of blocks that tend to be read or written together in a given environment is the first step towards powerful techniques for device failure isolation and power management. For example, identified groups can be placed together on a single disk, avoiding excess drive activity across an exascale storage system. Unlike previous grouping work, we focus on identifying groupings in data that can be gathered from real, running systems with minimal impact. Using temporal, spatial, and access ordering information from an enterprise data set, we identified a set of groupings that consistently appear, indicating that these are working sets that are likely to be accessed together. We present several techniques to obtain groupings along with a discussion of what techniques best apply to particular types of real systems. We intend to use these preliminary results to inform our search for new types of workloads with a goal of identifying properties of easily separable workloads across different systems and dynamically moving groups in these workloads to reduce disk activity in large storage systems.

BTRFS: The Linux B-Tree Filesystem BTRFS is a Linux filesystem that has been adopted as the default filesystem in some popular versions of Linux. It is based on copy-on-write, allowing for efficient snapshots and clones. It uses B-trees as its main on-disk data structure. The design goal is to work well for many use cases and workloads. To this end, much effort has been directed to maintaining even performance as the filesystem ages, rather than trying to support a particular narrow benchmark use-case. Linux filesystems are installed on smartphones as well as enterprise servers. This entails challenges on many different fronts. ---Scalability. The filesystem must scale in many dimensions: disk space, memory, and CPUs. ---Data integrity. Losing data is not an option, and much effort is expended to safeguard the content. This includes checksums, metadata duplication, and RAID support built into the filesystem. ---Disk diversity. The system should work well with SSDs and hard disks. It is also expected to be able to use an array of different sized disks, which poses challenges to the RAID and striping mechanisms. This article describes the core ideas, data structures, and algorithms of this filesystem. It sheds light on the challenges posed by defragmentation in the presence of snapshots, and the tradeoffs required to maintain even performance in the face of a wide spectrum of workloads.

Improving restore speed for backup systems that use inline chunk-based deduplication Slow restoration due to chunk fragmentation is a serious problem facing inline chunk-based data deduplication systems: restore speeds for the most recent backup can drop orders of magnitude over the lifetime of a system. We study three techniques--increasing cache size, container capping, and using a forward assembly area-- for alleviating this problem. Container capping is an ingest-time operation that reduces chunk fragmentation at the cost of forfeiting some deduplication, while using a forward assembly area is a new restore-time caching and prefetching technique that exploits the perfect knowledge of future chunk accesses available when restoring a backup to reduce the amount of RAM required for a given level of caching at restore time. We show that using a larger cache per stream--we see continuing benefits even up to 8 GB--can produce up to a 5-16X improvement, that giving up as little as 8% deduplication with capping can yield a 2-6X improvement, and that using a forward assembly area is strictly superior to LRU, able to yield a 2-4X improvement while holding the RAM budget constant.

Predictive data grouping: Defining the bounds of energy and latency reduction through predictive data grouping and replication We demonstrate that predictive grouping is an effective mechanism for reducing disk arm movement, thereby simultaneously reducing energy consumption and data access latency. We further demonstrate that predictive grouping has untapped dramatic potential to further improve access performance and limit energy consumption. Data retrieval latencies are considered a major bottleneck, and with growing volumes of data and increased storage needs it is only growing in significance. Data storage infrastructure is therefore a growing consumer of energy at data-center scales, while the individual disk is already a significant concern for mobile computing (accounting for almost a third of a mobile system's energy demands). While improving responsiveness of storage subsystems and hence reducing latencies in data retrieval is often considered contradictory with efforts to reduce disk energy consumption, we demonstrate that predictive data grouping has the potential to simultaneously work towards both these goals. Predictive data grouping has advantages in its applicability compared to both prior approaches to reducing latencies and to reducing energy usage. For latencies, grouping can be performed opportunistically, thereby avoiding the serious performance penalties that can be incurred with prior applications of access prediction (such as predictive prefetching of data). For energy, we show how predictive grouping can even save energy use for an individual disk that is never idle. Predictive data grouping with effective replication results in a reduction of the overall mechanical movement required to retrieve data. We have built upon our detailed measurements of disk power consumption, and have estimated both the energy expended by a hard disk for its mechanical components, and that needed to move the disk arm. We have further compared, via simulation, three models of predictive grouping of on-disk data, including an optimal arrangement of data that is guaranteed to minimize disk arm movement. These experiments have allowed us to measure the limits of performance improvement achievable with optimal data grouping and replication strategies on a single device, and have further allowed us to demonstrate the potential of such schemes to reduce energy consumption of mechanical components by up to 70&percnt;.

WorkOut: I/O workload outsourcing for boosting RAID reconstruction performance User I/O intensity can significantly impact the performance of on-line RAID reconstruction due to contention for the shared disk bandwidth. Based on this observation, this paper proposes a novel scheme, called WorkOut (I/O Workload Outsourcing), to significantly boost RAID reconstruction performance. WorkOut effectively outsources all write requests and popular read requests originally targeted at the degraded RAID set to a surrogate RAID set during reconstruction. Our lightweight prototype implementation of WorkOut and extensive trace-driven and benchmark-driven experiments demonstrate that, compared with existing reconstruction approaches, WorkOut significantly speeds up both the total reconstruction time and the average user response time. Importantly, WorkOut is orthogonal to and can be easily incorporated into any existing reconstruction algorithms. Furthermore, it can be extended to improving the performance of other background support RAID tasks, such as re-synchronization and disk scrubbing.

The Multi-Queue Replacement Algorithm for Second Level Buffer Caches This paper reports our research results that improve second level buffer cache performance. Several previous studies have shown that a good single level cache replacement algorithm such as LRU does not work well with second level buffer caches. Second level buffer caches have different access pattern from first level buffer caches because Accesses to second level buffer caches are actually misses from first level buffer caches.The paper presents our study of second level buffer cache access patterns using four large traces from various servers. We also introduce a new second level buffer cache replacement algorithm called Multi-Queue (MQ). Our trace-driven simulation results show that MQ performs better than all seven tested alternatives. Our implementation on a real storage system validates these results.

A sufficient condition for backtrack-bounded search Backtrack search is often used to solve constraint satisfaction problems. A relationship involving the structure of the constraints is described that provides a bound on the backtracking required to advance deeper into the backtrack tree. This analysis leads to upper bounds on the effort required for solution of a class of constraint satisfaction problems. The solutions involve a combination of relaxation preprocessing and backtrack search. The bounds are expressed in terms of the structure of the constraint connections. Specifically, the effort is shown to have a bound exponential in the size of the largest biconnected component of the constraint graph, as opposed to the size of the graph as a whole.

Applications of circumscription to formalizing common-sense knowledge Abstract We present a new and more symmetric version of the circumscription method of nonmonotonic reasoning rst described in (McCarthy 1980) and some applications to formalizing common,sense knowledge. The applications in this paper are mostly based on minimizing the abnormality of dieren t aspects of various entities. Included are nonmonotonic treatments of is-a hierarchies, the unique names hypothesis, and the frame problem. The new circumscription may be called formula circumscription to distinguish it from the previously dened domain circumscription and predicate circumscription. A still more general formalism called prioritized circumscription is briey explored.

The Complexity of Global Constraints We study the computational complexity of reasoning with global constraints. We show that reasoning with such constraints is intractable in general. We then demonstrate how the same tools of computational com- plexity can be used in the design and analysis of spe- cific global constraints. In particular, we illustrate how computational complexity can be used to determine when a lesser level of local consistency should be en- forced, when decomposing constraints will lose prun- ing, and when combining constraints is tractable. We also show how the same tools can be used to study symmetry breaking, meta-constraints like the cardinal- ity constraint, and learning nogoods.

A perspective on assumption-based truth maintenance

Formations of vehicles in cyclic pursuit Abstract—Inspired by the so-called “bugs” problem from mathematics, we study the geometric formations of multivehicle systems under cyclic pursuit. First, we introduce the notion of cyclic pursuit by examining a system of identical linear agents in the plane. This idea is then extended to a system of wheeled vehicles, each subject to a single nonholonomic constraint (i.e., unicycles), which is the principal focus of this paper. The pursuit framework is particularly simple in that the,identical vehicles are ordered such that vehicle pursues vehicle modulo . In this paper, we assume each vehicle has the same constant forward speed. We show that the system’s equilibrium formations are generalized regular polygons and it is exposed how the multivehicle system’s global behavior can be shaped through appropriate controller gain assignments. We then study the local stability of these equilibrium polygons, revealing which formations are stable and which are not. Index Terms—Circulant matrices, cooperative control, multia-

The Performance of Parity Placements in Disk Arrays Due to recent advances in central processing unit (CPU) and memory system performance, input/output (I/O) systems are increasingly limiting the performance of modern computer systems. Redundant arrays of inexpensive disks (RAID) have been proposed to meet the impending I/O crisis. RAIDs substitute many small inexpensive disks for a few large expensive disks to provide higher performance, smaller footprints, and lower power consumption at a lower cost than the large expensive disks they replace. RAIDs provide high availability by using parity encoding of data to survive disk failures. It is shown that the way parity is distributed in a RAID has significant consequences for performance. The performances of eight different parity placements are investigated using simulation.

Representing the process semantics in the situation calculus This paper presents a formal method based on the high-level semantics of processes to reason about continuous change. With a case study we show how the semantics of processes can be integrated with the situation calculus. The soundness and completeness of situation calculus with respect to the process semantics are proven. Furthermore, the logical programming is implemented to support the semantics of processes with the situation calculus.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

BDR: A Balanced Data Redistribution scheme to accelerate the scaling process of XOR-based Triple Disk Failure Tolerant arrays In large scale data centers, with the increasing amount of user data, Triple Disk Failure Tolerant arrays (3DFTs) gain much popularity due to their high reliability and low monetary cost. With the development of cloud computing, scalability becomes a challenging issue for disk arrays like 3DFTs. Although previous solutions improves the efficiency of RAID scaling, they suffer many problems (high I/O overhead and long migration time) in 3DFTs. It is because that existing approaches have to cost plenty of migration I/Os on balancing the data distribution according to the complex layout of erasure codes. To address this problem, we propose a novel Balanced Data Redistribution scheme (BDR) to accelerate the scaling process, which can be applied on XOR-based 3DFTs. BDR migrates proper data blocks according to a global point of view on a stripe set, which guarantees uniform data distribution and a small number of data movements. To demonstrate the effectiveness of BDR, we conduct several evaluations and simulations. The results show that, compared to typical RAID scaling approaches like Round-Robin (RR), SDM and RS6, BDR reduces the scaling I/Os by up to 77.45%, which speeds up the scaling process of 3DFTs by up to 4.17×, 3.31×, 3.88×, respectively.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Efficient parity placement schemes for tolerating up to two disk failures in disk arrays In order to achieve high reliability in disk array systems, two new schemes using dual parity placement, called DH1 (diagonal–horizontal) and DH2 schemes, are presented. Both DH schemes can tolerate up to two disk failures by using two types of parity information placed in the diagonal and the horizontal directions, respectively, in a matrix of disk partitions. DH1 scheme can reduce the occurrences of the bottleneck problem significantly because the parity blocks are evenly distributed throughout the disk array. DH2 scheme uses one more disk than DH1 scheme in order to store the horizontal parities, while the diagonal parities are placed in the same way as in DH1 scheme with a minor change. Even though both DH schemes use almost optimal disk space for storing the redundant information, the encoding algorithms for them are quite simple and efficient. Moreover, both DH schemes can recover rapidly from any two disk failures.

Bus Modelling in Zoned Disks RAID Storage Systems. A model of bus contention in a Multi-RAID storage architecture is presented. Based on an M/G/1 queue, the main issues are to determine the service time distribution that accurately represents the highly mixed input traffic of requests. This arises from the coexistence of different RAID organisations that generate several types of physical request (read/write for each RAID level) with different related sizes. The size distributions themselves are made more complex by the striping mechanism, with full/large/small stripes in RAID5. We show the impact of the bus traffic on the system's overall performance as predicted by the model and validated against a simulation of the hardware, using common workload assumptions.

RAID level selection for heterogeneous disk arrays Heterogeneous Disk Arrays (HDAs) allow resource sharing of their hardware by multiple RAID levels. RAID1 (mirrored disks) and RAID5 (distributed parity arrays) are the two RAID levels considered in this study. They are both single disk failure tolerant (1DFT), but differ significantly in their efficiency in processing database workloads. The goal of the study is to maximize the number of Virtual Array (VA) allocations in HDA. We develop an analysis to estimate the load per VA based on a few parameters: the fraction of accesses to small versus large blocks and the fraction of updates versus reads. A VA is allocated according to the RAID level, which minimizes the anticipated load based on input parameters. Operation in normal and degraded mode is considered for comparison purposes, but in fact allocations are carried out using the higher load in degraded mode to ensure that single disk failures will not result in overload. We report on parametric studies to gain insight into circumstances leading to a RAID1 or RAID5 classification. An allocation experiment with a synthetic workload is used to demonstrate the superiority of HDA with respect to purely RAID1 or RAID5 disk arrays. This analytic study can be extended to 2DFT arrays, namely RAID6 versus 3-way replication.

A multiple disk failure recovery scheme in RAID systems In this paper, we propose a practical disk error recovery scheme tolerating multiple simultaneous disk failures in a typical RAID system, resulting in improvement in availability and reliability. The scheme is composed of the encoding and the decoding processes. The encoding process is defined by making one horizontal parity and a number of vertical parities. The decoding process is defined by a data recovering method for multiple disk failures including the parity disks. The proposed error recovery scheme is proven to correctly recover the original data for multiple simultaneous disk failures regardless of the positions of the failed disks. The proposed error recovery scheme only uses exclusive OR operations and simple arithmetic operations, which can be easily implemented on current RAID systems without hardware changes.

A Highly Accurate Method for Assessing Reliability of Redundant Arrays of Inexpensive Disks (RAID) Abstract - The statistical bases for current models of RAID reliability are reviewed and a highly accurate alternative is provided and justified. This new model corrects statistical errors associated with the pervasive assumption that system (RAID group) times to failure follow a homogeneous Poisson process, and corrects errors associated with assuming the time-to-failure and time-to-restore distributions are exponentially distributed. Statistical justification for the new model uses theory for reliability of repairable systems. Four critical component distributions are developed from field data. These distributions are for times to catastrophic failure, reconstruction and restoration, read errors, and disk data scrubs. Model results have been verified and predict between 2 to 1,500 times as many double disk failures as estimates made using the mean time to data loss method. Model results are compared to system level field data for RAID group of 14 drives and show excellent correlation and greater accuracy than either MTTDL.

Serverless network file systems We propose a new paradigm for network file system design: serverless network file systems. While traditional network file systems rely on a central server machine, a serverless system utilizes workstations cooperating as peers to provide all file system services. Any machine in the system can store, cache, or control any block of data. Our approach uses this location independence, in combination with fast local area networks, to provide better performance and scalability than traditional file systems. Furthermore, because any machine in the system can assume the responsibilities of a failed component, our serverless design also provides high availability via redundatn data storage. To demonstrate our approach, we have implemented a prototype serverless network file system called xFS. Preliminary performance measurements suggest that our architecture achieves its goal of scalability. For instance, in a 32-node xFS system with 32 active clients, each client receives nearly as much read or write throughput as it would see if it were the only active client.

Accurate and efficient replaying of file system traces Years of innovation in file systems have been highly successful in improving their performance and functionality, but at the cost of complicating their interaction with the disk. A variety of techniques exist to ensure consistency and integrity of file ...

Choosing the best storage system for video service

Performance study of RAID-5 disk arrays with data and parity cache Disk array architectures such as RAID-5 have become an acceptable way far designing highly reliable and high-performance storage systems. However one major drawback of a RAID-5 disk array system is that an update to a data block may involve four disk accesses. Such a high overhead is especially undesirable for workloads with high update rate as in transaction processing. In this paper, we present a new scheme for improving the write performance of disk arrays using controller cache to store data as well as parity information. We have developed a trace-driven model to simulate cached disk arrays for transaction processing environment. We have studied the effect of caching parity information at the controller level along with caching data. The simulation results show a considerable improvement in response time of data and parity cached disk array over disk arrays with only data caching. The improvement in response time for disk array employing parity cache is about 10%-20% for the parameters used in our study

A Mechanism for Managing the Buffer Pool in a Relational Database System Using the Hot Set Model

Worlds to die for We last had an "open problems" column eighteen months ago [Hem94]. It contained seven problems. Of the seven, one has since been resolved (at least insofar as one can resolve the problem without outright collapsing complexity classes) in an exciting FOCS paper by Cai and Sivakumar ([CS95], see also [Ogi95b,CNS95]), and for another I received a proof via email unfortunately followed quickly by another email retracting the proof. Overall score:Mysteries of Complexity Theory: 6.Theoretical Computer Scientists:1.If you go to Atlantic City, you know which side to bet on! But be of good cheer. This issue's column contains a new list of open problems (though some favorites from the old list have stowed away here too). And to stack the deck in favor of theoretical computer scientists, the problems are posed quite obliquely. Rather than asking you to prove "X," many of the problems (e.g., Problems 2, 4, 5, 6, and 7) just ask you to show that "In some oracle world, X." Sound easy? Dig in! And if your attempt to find a world where X holds becomes too frustrating, don't hesitate to go for the real glory --- by proving that X fails in the real world (and every relativized world)!

Contingent planning with goal preferences The importance of the problems of contingent planning with actions that have non-deterministic effects and of planning with goal preferences has been widely recognized, and several works address these two problems separately. However, combining conditional planning with goal preferences adds some new difficulties to the problem. Indeed, even the notion of optimal plan is far from trivial, since plans in nondeterministic domains can result in several different behaviors satisfying conditions with different preferences. Planning for optimal conditional plans must therefore take into account the different behaviors, and conditionally search for the highest preference that can be achieved. In this paper, we address this problem. We formalize the notion of optimal conditional plan, and we describe a correct and complete planning algorithm that is guaranteed to find optimal solutions. We implement the algorithm using BDD-based techniques, and show the practical potentialities of our approach through a preliminary experimental evaluation.

Relating equivalence and reducibility to sparse sets For various polynomial-time reducibilities r, the authors ask whether being r-reducible to a sparse set is a broader notion than being r-equivalent to a sparse set. Although distinguishing equivalence and reducibility to sparse sets, for many-one or 1-truth-table reductions, would imply that P≠NP, the authors show that for k-truth-table reductions, k⩾2, equivalence and reducibility to sparse sets provably differ. Though R. Gavalda and D. Watanabe have shown that, for any polynomial-time computable unbounded function f(·), some sets f(n)-truth-table reducible to sparse sets are not even Turing equivalent to sparse sets, the authors show that extending their result to the 2-truth-table case would provide a proof that P≠NP. Additionally, the authors study the relative power of different notions of reducibility and show that disjunctive and conjunctive truth-table reductions to sparse sets are surprisingly powerful, refuting a conjecture of K. Ko (1989)

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Mutual Information Regularized Feature-Level Frankenstein for Discriminative Recognition Deep learning recognition approaches can potentially perform better if we can extract a discriminative representation that controllably separates nuisance factors. In this paper, we propose a novel approach to explicitly enforce the extracted discriminative representation <inline-formula><tex-math notation="LaTeX">$\boldsymbol{d}$</tex-math></inline-formula> , extracted latent variation <inline-formula><tex-math notation="LaTeX">$\boldsymbol{l}$</tex-math></inline-formula> (e,g., background, unlabeled nuisance attributes), and semantic variation label vector <inline-formula><tex-math notation="LaTeX">$\boldsymbol{s}$</tex-math></inline-formula> (e.g., labeled expressions/pose) to be independent and complementary to each other. We can cast this problem as an adversarial game in the latent space of an auto-encoder. Specifically, with the to-be-disentangled <inline-formula><tex-math notation="LaTeX">$\boldsymbol{s}$</tex-math></inline-formula> , we propose to equip an end-to-end conditional adversarial network with the ability to decompose an input sample into <inline-formula><tex-math notation="LaTeX">${\boldsymbol{d}}$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$\boldsymbol{l}$</tex-math></inline-formula> . However, we argue that maximizing the cross-entropy loss of semantic variation prediction from <inline-formula><tex-math notation="LaTeX">$\boldsymbol{d}$</tex-math></inline-formula> is not sufficient to remove the impact of <inline-formula><tex-math notation="LaTeX">$\boldsymbol{s}$</tex-math></inline-formula> from <inline-formula><tex-math notation="LaTeX">$\boldsymbol{d}$</tex-math></inline-formula> , and that the uniform-target and entropy regularization are necessary. A collaborative mutual information regularization framework is further proposed to avoid unstable adversarial training. It is able to minimize the differentiable mutual information between the variables to enforce independence. The proposed discriminative representation inherits the desired tolerance property guided by prior knowledge of the task. Our proposed framework achieves top performance on diverse recognition tasks, including digits classification, large-scale face recognition on LFW and IJB-A datasets, and face recognition tolerant to changes in lighting, makeup, disguise, etc.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Assessing the Expressivity of Planning Formalisms through the Comparison to Formal Languages. From a theoretical perspective, judging the expressivity of planning formalisms helps to understand the relationship of different representations and to infer theoretical properties. From a practical point of view, it is important to be able to choose the best formalism for a problem at hand, or to ponder the consequences of introducing new representation features. Most work on the expressivity is based either on compilation approaches, or on the computational complexity of the plan existence problem. Recently, we introduced a new notion of expressivity. It is based on comparing the structural complexity of the set of solutions to a planning problem by interpreting the set as a formal language and classifying it with respect to the Chomsky hierarchy. This is a more direct measure than the plan existence problem and enables also the comparison of formalisms that can not be compiled into each other. While existing work on that last approach focused on different hierarchical problem classes, this paper investigates STRIPS with and without conditional effects; though we also tighten some existing results on hierarchical formalisms. Our second contribution is a discussion on the language-based expressivity measure with respect to the other approaches.

Tight Bounds for HTN Planning. Although HTN planning is in general undecidable, there are many syntactically identifiable sub-classes of HTN problems that can be decided. For these sub-classes, the decision procedures provide upper complexity bounds. Lower bounds were often not investigated in more detail, however. We generalize a propositional HTN formalization to one that is based upon a function-free first-order logic and provide tight upper and lower complexity results along three axes: whether variables are allowed in operator and method schemas, whether the initial task and methods must be totally ordered, and where recursion is allowed (arbitrary recursion, tail-recursion, and acyclic problems). Our findings have practical implications, both for the reuse of classical planning techniques for HTN planning, and for the design of efficient HTN algorithms.

On the decidability of HTN planning with task insertion The field of deterministic AI planning can roughly be divided into two approaches -- classical state-based planning and hierarchical task network (HTN) planning. The plan existence problem of the former is known to be decidable while it has been proved undecidable for the latter. When extending HTN planning by allowing the unrestricted insertion of tasks and ordering constraints, one obtains a form of planning which is often referred to as "hybrid planning". We present a simplified formalization of HTN planning with and without task insertion. We show that the plan existence problem is undecidable for the HTN setting without task insertion and that it becomes decidable when allowing task insertion. In the course of the proof, we obtain an upper complexity bound of EXPSPACE for the plan existence problem for propositional HTN planning with task insertion.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Logic Programming and Negation: A Survey. We survey here various approaches which were proposed to incorporate negation in logicprograms. We concentrate on the proof-theoretic and model-theoretic issues and the relationshipsbetween them.1991 Mathematics Subject Classification: 68Q40, 68T15.CR Categories: F.3.2., F.4.1, H.3.3, I.2.3.Keywords and Phrases: negation, general logic programs, non-monotonic reasoning.Notes. The work of the first author was partly supported by ESPRIT Basic Research Action6810 (Compulog 2). The work...

A sufficient condition for backtrack-bounded search Backtrack search is often used to solve constraint satisfaction problems. A relationship involving the structure of the constraints is described that provides a bound on the backtracking required to advance deeper into the backtrack tree. This analysis leads to upper bounds on the effort required for solution of a class of constraint satisfaction problems. The solutions involve a combination of relaxation preprocessing and backtrack search. The bounds are expressed in terms of the structure of the constraint connections. Specifically, the effort is shown to have a bound exponential in the size of the largest biconnected component of the constraint graph, as opposed to the size of the graph as a whole.

Convergence of a Nonconforming Multiscale Finite Element Method The multiscale finite element method (MsFEM) [T. Y. Hou, X. H. Wu, and Z. Cai, Math. Comp., 1998, to appear; T. Y. Hou and X. H. Wu, J. Comput. Phys., 134 (1997), pp. 169--189] has been introduced to capture the large scale solutions of elliptic equations with highly oscillatory coefficients. This is accomplished by constructing the multiscale base functions from the local solutions of the elliptic operator. Our previous study reveals that the leading order error in this approach is caused by the ``resonant sampling,'' which leads to large error when the mesh size is close to the small scale of the continuous problem. Similar difficulty also arises in numerical upscaling methods. An oversampling technique has been introduced to alleviate this difficulty [T. Y. Hou and X. H. Wu, J. Comput. Phys., 134 (1997), pp. 169--189]. A consequence of the oversampling method is that the resulting finite element method is no longer conforming. Here we give a detailed analysis of the nonconforming error. Our analysis also reveals a new cell resonance error which is caused by the mismatch between the mesh size and the wavelength of the small scale. We show that the cell resonance error is of lower order. Our numerical experiments demonstrate that the cell resonance error is generically small and is difficult to observe in practice.

Efficient sparse coding algorithms Sparse coding provides a class of algorithms for finding succinct representations of stimuli; given only unlabeled input data, it discovers basis functions that cap- ture higher-level features in the data. However, finding sparse codes remains a very difficult computational problem. In this paper, we present efficient sparse coding algorithms that are based on iteratively solving two convex optimization problems: an L1-regularized least squares problem and an L2-constrained least squares problem. We propose novel algorithms to solve both of these optimiza- tion problems. Our algorithms result in a significant speedup for sparse coding, allowing us to learn larger sparse codes than possible with previously described algorithms. We apply these algorithms to natural images and demonstrate that the inferred sparse codes exhibit end-stopping and non-classical receptive field sur- round suppression and, therefore, may provide a partial explanation for these two phenomena in V1 neurons.

Synchronized Disk Interleaving A group of disks may be interleaved to speed up data transfers in a manner analogous to the speedup achieved by main memory interleaving. Conventional disks may be used for interleaving by spreading data across disks and by treating multiple disks as if they were a single one. Furthermore, the rotation of the interleaved disks may be synchronized to simplify control and also to optimize performance. In addition, check- sums may be placed on separate check-sum disks in order to improve reliability. In this paper, we study synchronized disk interleaving as a high-performance mass storage system architecture. The advantages and limitations of the proposed disk interleaving scheme are analyzed using the M/G/1 queueing model and compared to the conventional disk access mechanism.

A Completeness Result for SLDNF-Resolution Because of the possibility of floundering and infinite derivations, SLDNF-resolution is, in general, not complete. The classical approach [17] to get a completeness result is to restrict the attention to normal programs P and normal goals G, such that P or {G} is allowed and P is hierarchical. Unfortunately, the class of all normal programs and all normal goals meeting these requirements is not powerful enough to be of great practical importance. But after refining the concept of allowedness by taking modes [12] into account, we can broaden the notion of a hierarchical program, and thereby define a subclass of the class of normal programs and normal goals which is powerful enough to compute all primitive recursive functions without losing the completeness of SLDNF-resolution.

Diagnostic reasoning with A-Prolog In this paper, we suggest an architecture for a software agent which operates a physical device and is capable of making observations and of testing and repairing the device's components. We present simplified definitions of the notions of symptom, candidate diagnosis, and diagnosis which are based on the theory of action language ${\cal AL}$. The definitions allow one to give a simple account of the agent's behavior in which many of the agent's tasks are reduced to computing stable models of logic programs.

ARIMA time series modeling and forecasting for adaptive I/O prefetching Bursty application I/O patterns, together with transfer limited storage devices, combine to create a major I/O bottleneck on parallel systems. This paper explores the use of time series models to forecast application I/O request times, then prefetching I/O requests during computation intervals to hide I/O latency. Experimental results with I/O intensive scientific codes show performance improvements compared to standard UNIX prefetching strategies.

Scheduling parallel I/O operations The I/O bottleneck in parallel computer systems has recently begun receiving increasing interest. Most attention has focused on improving the performance of I/O devices using fairly low-level parallelism in techniques such as disk striping and interleaving. Widely applicable solutions, however, will require an integrated approach which addresses the problem at multiple system levels, including applications, systems software, and architecture. We propose that within the context of such an integrated approach, scheduling parallel I/O operations will become increasingly attractive and can potentially provide substantial performance benefits.We describe a simple I/O scheduling problem and present approximate algorithms for its solution. The costs of using these algorithms in terms of execution time, and the benefits in terms of reduced time to complete a batch of I/O operations, are compared with the situations in which no scheduling is used, and in which an optimal scheduling algorithm is used. The comparison is performed both theoretically and experimentally. We have found that, in exchange for a small execution time overhead, the approximate scheduling algorithms can provide substantial improvements in I/O completion times.

Learning A Lexical Simplifier Using Wikipedia In this paper we introduce a new lexical simplification approach. We extract over 30K candidate lexical simplifications by identifying aligned words in a sentence-aligned corpus of English Wikipedia with Simple English Wikipedia. To apply these rules, we learn a feature-based ranker using SVMnk trained on a set of labeled simplifications collected using Amazon's Mechanical Turk. Using human simplifications for evaluation, we achieve a precision of 76% with changes in 86% of the examples.

Trends in multiprocessor and distributed operating systems designs This paper presents an overview of the developments in operating systems technology for distributed computing systems and multiprocessor machines. We focus on those design principles that are now widely accepted as useful design paradigms. Approaches common to distributed and multiprocessor operating systems are identified. Issues discussed include process scheduling and synchronization, load balancing, virtual and shared-memory management and parallel file systems. The task-thread model and the object model of computing are also discussed.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Greedy Part-Wise Learning of Sum-Product Networks.

Multimodal Transitions for Generative Stochastic Networks. Generative Stochastic Networks (GSNs) have been recently introduced as an alternative to traditional probabilistic modeling: instead of parametrizing the data distribution directly, one parametrizes a transition operator for a Markov chain whose stationary distribution is an estimator of the data generating distribution. The result of training is therefore a machine that generates samples through this Markov chain. However, the previously introduced GSN consistency theorems suggest that in order to capture a wide class of distributions, the transition operator in general should be multimodal, something that has not been done before this paper. We introduce for the first time multimodal transition distributions for GSNs, in particular using models in the NADE family (Neural Autoregressive Density Estimator) as output distributions of the transition operator. A NADE model is related to an RBM (and can thus model multimodal distributions) but its likelihood (and likelihood gradient) can be computed easily. The parameters of the NADE are obtained as a learned function of the previous state of the learned Markov chain. Experiments clearly illustrate the advantage of such multimodal transition distributions over unimodal GSNs.

Representation Models In Single Channel Source Separation Model-based single-channel source separation (SCSS) is an ill-posed problem requiring source-specific prior knowledge. In this paper, we use representation learning and compare general stochastic networks (GSNs), Gauss Bernoulli restricted Boltzmann machines (GBRBMs), conditional Gauss Bernoulli restricted Boltzmann machines (CGBRBMs), and higher order contractive autoencoders (HCAEs) for modeling the source-specific knowledge. In particular, these models learn a mapping from speech mixture spectrogram representations to single-source spectrogram representations, i.e. we apply them as filter for the speech mixture. In the test case, the individual source spectrograms of both models are inferred and the softmask for re-synthesis of the time signals is determined thereof. We evaluate the deep architectures on data of the 2nd CHiME speech separation challenge and provide results for a speaker dependent, a speaker independent, a matched noise condition and an unmatched noise condition task. Our experiments show the best PESQ and overall perceptual score on average for GSNs in all four tasks.

Regularized Auto-Encoders Estimate Local Statistics

A regression approach to speech enhancement based on deep neural networks In contrast to the conventional minimum mean square error (MMSE)-based noise reduction techniques, we propose a supervised method to enhance speech by means of finding a mapping function between noisy and clean speech signals based on deep neural networks (DNNs). In order to be able to handle a wide range of additive noises in real-world situations, a large training set that encompasses many possible combinations of speech and noise types, is first designed. A DNN architecture is then employed as a nonlinear regression function to ensure a powerful modeling capability. Several techniques have also been proposed to improve the DNN-based speech enhancement system, including global variance equalization to alleviate the over-smoothing problem of the regression model, and the dropout and noise-aware training strategies to further improve the generalization capability of DNNs to unseen noise conditions. Experimental results demonstrate that the proposed framework can achieve significant improvements in both objective and subjective measures over the conventional MMSE based technique. It is also interesting to observe that the proposed DNN approach can well suppress highly nonstationary noise, which is tough to handle in general. Furthermore, the resulting DNN model, trained with artificial synthesized data, is also effective in dealing with noisy speech data recorded in real-world scenarios without the generation of the annoying musical artifact commonly observed in conventional enhancement methods.

Efficient sparse coding algorithms Sparse coding provides a class of algorithms for finding succinct representations of stimuli; given only unlabeled input data, it discovers basis functions that cap- ture higher-level features in the data. However, finding sparse codes remains a very difficult computational problem. In this paper, we present efficient sparse coding algorithms that are based on iteratively solving two convex optimization problems: an L1-regularized least squares problem and an L2-constrained least squares problem. We propose novel algorithms to solve both of these optimiza- tion problems. Our algorithms result in a significant speedup for sparse coding, allowing us to learn larger sparse codes than possible with previously described algorithms. We apply these algorithms to natural images and demonstrate that the inferred sparse codes exhibit end-stopping and non-classical receptive field sur- round suppression and, therefore, may provide a partial explanation for these two phenomena in V1 neurons.

Exploring Strategies for Training Deep Neural Networks Deep multi-layer neural networks have many levels of non-linearities allowing them to compactly represent highly non-linear and highly-varying functions. However, until recently it was not clear how to train such deep networks, since gradient-based optimization starting from random initialization often appears to get stuck in poor solutions. Hinton et al. recently proposed a greedy layer-wise unsupervised learning procedure relying on the training algorithm of restricted Boltzmann machines (RBM) to initialize the parameters of a deep belief network (DBN), a generative model with many layers of hidden causal variables. This was followed by the proposal of another greedy layer-wise procedure, relying on the usage of autoassociator networks. In the context of the above optimization problem, we study these algorithms empirically to better understand their success. Our experiments confirm the hypothesis that the greedy layer-wise unsupervised training strategy helps the optimization by initializing weights in a region near a good local minimum, but also implicitly acts as a sort of regularization that brings better generalization and encourages internal distributed representations that are high-level abstractions of the input. We also present a series of experiments aimed at evaluating the link between the performance of deep neural networks and practical aspects of their topology, for example, demonstrating cases where the addition of more depth helps. Finally, we empirically explore simple variants of these training algorithms, such as the use of different RBM input unit distributions, a simple way of combining gradient estimators to improve performance, as well as on-line versions of those algorithms.

Unsupervised Learning of Models for Recognition We present a method to learn object class models from unlabeled and unsegmented cluttered scenes for the purpose of visual object recognition. We focus on a particular type of model where objects are represented as flexible constellations of rigid parts (features). The variability within a class is represented by a joint probability density function (pdf) on the shape of the constellation and the output of part detectors. In a first stage, the method automatically identifies distinctive parts in the training set by applying a clustering algorithm to patterns selected by an interest operator. It then learns the statistical shape model using expectation maximization. The method achieves very good classification results on human faces and rear views of cars.

Semi-supervised learning of compact document representations with deep networks Finding good representations of text documents is crucial in information retrieval and classification systems. Today the most popular document representation is based on a vector of word counts in the document. This representation neither captures dependencies between related words, nor handles synonyms or polysemous words. In this paper, we propose an algorithm to learn text document representations based on semi-supervised autoencoders that are stacked to form a deep network. The model can be trained efficiently on partially labeled corpora, producing very compact representations of documents, while retaining as much class information and joint word statistics as possible. We show that it is advantageous to exploit even a few labeled samples during training.

Proceedings of the 17th ACM Conference on Information and Knowledge Management, CIKM 2008, Napa Valley, California, USA, October 26-30, 2008

PatternHunter II: highly sensitive and fast homology search. Extending the single optimized spaced seed of PatternHunter to multiple ones, PatternHunter II simultaneously remedies the lack of sensitivity of Blastn and the lack of speed of Smith-Waterman, for homology search. At Blastn speed, PatternHunter II approaches Smith-Waterman sensitivity, bringing homology search technology back to a full circle.

A new approach to I/O performance evaluation: self-scaling I/O benchmarks, predicted I/O performance Current I/O benchmarks suffer from several chronic problems: they quickly become obsolete, they do not stress the I/O system, and they do not help in understanding I/O system performance. We propose a new approach to I/O performance analysis. First, we propose a self-scaling benchmark that dynamically adjusts aspects of its workload according to the performance characteristic of the system being measured. By doing so, the benchmark automatically scales across current and future systems. The evaluation aids in understanding system performance by reporting how performance varies according to each of fie workload parameters. Second, we propose predicted performance, a technique for using the results from the self-scaling evaluation to quickly estimate the performance for workloads that have not been measured. We show that this technique yields reasonably accurate performance estimates and argue that this method gives a far more accurate comparative performance evaluation than traditional single point benchmarks. We apply our new evaluation technique by measuring a SPARCstation 1+ with one SCSI disk, an HP 730 with one SCSI-II disk, a Sprite LFS DECstation 5000/200 with a three-disk disk array, a Convex C240 minisupercomputer with a four-disk disk array, and a Solbourne 5E/905 fileserver with a two-disk disk array.

Exploiting Web Log Mining for Web Cache Enhancement Improving the performance of the Web is a crucial requirement, since its popularity resulted in a large increase in the user perceived latency. In this paper, we describe a Web caching scheme that capitalizes on prefetching. Prefetching refers to the mechanism of deducing forthcoming page accesses of a client, based on access log information. Web log mining methods are exploited to provide effective prediction of Web-user accesses. The proposed scheme achieves a coordination between the two techniques (i.e., caching and prefetching). The prefetched documents are accommodated in a dedicated part of the cache, to avoid the drawback of incorrect replacement of requested documents. The requirements of the Web are taken into account, compared to the existing schemes for buffer management in database and operating systems. Experimental results indicate the superiority of the proposed method compared to the previous ones, in terms of improvement in cache performance.

Learning A Lexical Simplifier Using Wikipedia In this paper we introduce a new lexical simplification approach. We extract over 30K candidate lexical simplifications by identifying aligned words in a sentence-aligned corpus of English Wikipedia with Simple English Wikipedia. To apply these rules, we learn a feature-based ranker using SVMnk trained on a set of labeled simplifications collected using Amazon's Mechanical Turk. Using human simplifications for evaluation, we achieve a precision of 76% with changes in 86% of the examples.

Data-intensive applications, challenges, techniques and technologies: A survey on Big Data. It is already true that Big Data has drawn huge attention from researchers in information sciences, policy and decision makers in governments and enterprises. As the speed of information growth exceeds Moore’s Law at the beginning of this new century, excessive data is making great troubles to human beings. However, there are so much potential and highly useful values hidden in the huge volume of data. A new scientific paradigm is born as data-intensive scientific discovery (DISD), also known as Big Data problems. A large number of fields and sectors, ranging from economic and business activities to public administration, from national security to scientific researches in many areas, involve with Big Data problems. On the one hand, Big Data is extremely valuable to produce productivity in businesses and evolutionary breakthroughs in scientific disciplines, which give us a lot of opportunities to make great progresses in many fields. There is no doubt that the future competitions in business productivity and technologies will surely converge into the Big Data explorations. On the other hand, Big Data also arises with many challenges, such as difficulties in data capture, data storage, data analysis and data visualization. This paper is aimed to demonstrate a close-up view about Big Data, including Big Data applications, Big Data opportunities and challenges, as well as the state-of-the-art techniques and technologies we currently adopt to deal with the Big Data problems. We also discuss several underlying methodologies to handle the data deluge, for example, granular computing, cloud computing, bio-inspired computing, and quantum computing.

Gaussian-binary Restricted Boltzmann Machines on Modeling Natural Image Statistics. We present a theoretical analysis of Gaussian-binary restricted Boltzmann machines (GRBMs) from the perspective of density models. The key aspect of this analysis is to show that GRBMs can be formulated as a constrained mixture of Gaussians, which gives a much better insight into the model's capabilities and limitations. We show that GRBMs are capable of learning meaningful features both in a two-dimensional blind source separation task and in modeling natural images. Further, we show that reported difficulties in training GRBMs are due to the failure of the training algorithm rather than the model itself. Based on our analysis we are able to propose several training recipes, which allowed successful and fast training in our experiments. Finally, we discuss the relationship of GRBMs to several modifications that have been proposed to improve the model.

De novo identification of replication-timing domains in the human genome by deep learning. Motivation: The de novo identification of the initiation and termination zones-regions that replicate earlier or later than their upstream and downstream neighbours, respectively-remains a key challenge in DNA replication. Results: Building on advances in deep learning, we developed a novel hybrid architecture combining a pre-trained, deep neural network and a hidden Markov model (DNN-HMM) for the de novo identification of replication domains using replication timing profiles. Our results demonstrate that DNN-HMM can significantly outperform strong, discriminatively trained Gaussian mixture model-HMM (GMM-HMM) systems and other six reported methods that can be applied to this challenge. We applied our trained DNN-HMM to identify distinct replication domain types, namely the early replication domain (ERD), the down transition zone (DTZ), the late replication domain (LRD) and the up transition zone (UTZ), using newly replicated DNA sequencing (Repli-Seq) data across 15 human cells. A subsequent integrative analysis revealed that these replication domains harbour unique genomic and epigenetic patterns, transcriptional activity and higher-order chromosomal structure. Our findings support the 'replication-domain' model, which states (1) that ERDs and LRDs, connected by UTZs and DTZs, are spatially compartmentalized structural and functional units of higher-order chromosomal structure, (2) that the adjacent DTZ-UTZ pairs form chromatin loops and (3) that intra-interactions within ERDs and LRDs tend to be short-range and long-range, respectively. Our model reveals an important chromatin organizational principle of the human genome and represents a critical step towards understanding the mechanisms regulating replication timing.

Trends in extreme learning machines: A review. Extreme learning machine (ELM) has gained increasing interest from various research fields recently. In this review, we aim to report the current state of the theoretical research and practical advances on this subject. We first give an overview of ELM from the theoretical perspective, including the interpolation theory, universal approximation capability, and generalization ability. Then we focus on the various improvements made to ELM which further improve its stability, sparsity and accuracy under general or specific conditions. Apart from classification and regression, ELM has recently been extended for clustering, feature selection, representational learning and many other learning tasks. These newly emerging algorithms greatly expand the applications of ELM. From implementation aspect, hardware implementation and parallel computation techniques have substantially sped up the training of ELM, making it feasible for big data processing and real-time reasoning. Due to its remarkable efficiency, simplicity, and impressive generalization performance, ELM have been applied in a variety of domains, such as biomedical engineering, computer vision, system identification, and control and robotics. In this review, we try to provide a comprehensive view of these advances in ELM together with its future perspectives.

Learning deep representations via extreme learning machines. Extreme learning machine (ELM) as an emerging technology has achieved exceptional performance in large-scale settings, and is well suited to binary and multi-class classification, as well as regression tasks. However, existing ELM and its variants predominantly employ single hidden layer feedforward networks, leaving the popular and potentially powerful stacked generalization principle unexploited for seeking predictive deep representations of input data. Deep architectures can find higher-level representations, thus can potentially capture relevant higher-level abstractions. But most of current deep learning methods require solving a difficult and non-convex optimization problem. In this paper, we propose a stacked model, DrELM, to learn deep representations via extreme learning machine according to stacked generalization philosophy. The proposed model utilizes ELM as a base building block and incorporates random shift and kernelization as stacking elements. Specifically, in each layer, DrELM integrates a random projection of the predictions obtained by ELM into the original feature, and then applies kernel functions to generate the resultant feature. To verify the classification and regression performance of DrELM, we conduct the experiments on both synthetic and real-world data sets. The experimental results show that DrELM outperforms ELM and kernel ELMs, which appear to demonstrate that DrELM could yield predictive features that are suitable for prediction tasks. The performances of the deep models (i.e. Stacked Auto-encoder) are comparable. However, due to the utilization of ELM, DrELM is easier to learn and faster in testing.

Modeling Human Motion Using Binary Latent Variables We propose a non-linear generative model for human motion data that uses an undirected model with binary latent variables and real-valued "visible" variables that represent joint angles. The latent and visible variabl es at each time step re- ceive directed connections from the visible variables at th e last few time-steps. Such an architecture makes on-line inference efficient and a llows us to use a sim- ple approximate learning procedure. After training, the model finds a single set of parameters that simultaneously capture several differe nt kinds of motion. We demonstrate the power of our approach by synthesizing various motion sequences and by performing on-line filling in of data lost during motio n capture. Website: http://www.cs.toronto.edu/∼gwtaylor/publications/nips2006mhmublv/

Traffic Flow Prediction With Big Data: A Deep Learning Approach Accurate and timely traffic flow information is important for the successful deployment of intelligent transportation systems. Over the last few years, traffic data have been exploding, and we have truly entered the era of big data for transportation. Existing traffic flow prediction methods mainly use shallow traffic prediction models and are still unsatisfying for many real-world applications. This situation inspires us to rethink the traffic flow prediction problem based on deep architecture models with big traffic data. In this paper, a novel deep-learning-based traffic flow prediction method is proposed, which considers the spatial and temporal correlations inherently. A stacked autoencoder model is used to learn generic traffic flow features, and it is trained in a greedy layerwise fashion. To the best of our knowledge, this is the first time that a deep architecture model is applied using autoencoders as building blocks to represent traffic flow features for prediction. Moreover, experiments demonstrate that the proposed method for traffic flow prediction has superior performance.

Sum-Product Networks: A New Deep Architecture The key limiting factor in graphical model inference and learning is the complexity of the partition function. We thus ask the question: what are general conditions under which the partition function is tractable? The answer leads to a new kind of deep architecture, which we call sum-product networks (SPNs). SPNs are directed acyclic graphs with variables as leaves, sums and products as internal nodes, and weighted edges. We show that if an SPN is complete and consistent it represents the partition function and all marginals of some graphical model, and give semantics to its nodes. Essentially all tractable graphical models can be cast as SPNs, but SPNs are also strictly more general. We then propose learning algorithms for SPNs, based on backpropagation and EM. Experiments show that inference and learning with SPNs can be both faster and more accurate than with standard deep networks. For example, SPNs perform image completion better than state-of-the-art deep networks for this task. SPNs also have intriguing potential connections to the architecture of the cortex.

Learning Features from Music Audio with Deep Belief Networks.

Training Hierarchical Feed-Forward Visual Recognition Models Using Transfer Learning from Pseudo-Tasks Building visual recognition models that adapt across different domains is a challenging task for computer vision. While feature-learning machines in the form of hierarchial feed-forward models (e.g., convolutional neural networks) showed promise in this direction, they are still difficult to train especially when few training examples are available. In this paper, we present a framework for training hierarchical feed-forward models for visual recognition, using transfer learning from pseudo tasks. These pseudo tasks are automatically constructed from data without supervision and comprise a set of simple pattern-matching operations. We show that these pseudo tasks induce an informative inverse-Wishart prior on the functional behavior of the network, offering an effective way to incorporate useful prior knowledge into the network training. In addition to being extremely simple to implement, and adaptable across different domains with little or no extra tuning, our approach achieves promising results on challenging visual recognition tasks, including object recognition, gender recognition, and ethnicity recognition.

Implementation and performance of integrated application-controlled file caching, prefetching, and disk scheduling As the performance gap between disks and micropocessors continues to increase, effective utilization of the file cache becomes increasingly immportant. Application-controlled file caching and prefetching can apply application-specific knowledge to improve file cache management. However, supporting application-controlled file caching and prefetching is nontrivial because caching and prefetching need to be integrated carefully, and the kernel needs to allocate cache blocks among processes appropriately. This article presents the design, implementation, and performance of a file system that integrates application-controlled caching, prefetching, and disk scheduling. We use a two-level cache management strategy. The kernel uses the LRU-SP (Least-Recently-Used with Swapping and Placeholders) policy to allocate blocks to processes, and each process integrates application-specific caching and prefetching based on the controlled-aggressive policy, an algorithm previously shown in a theoretical sense to be nearly optimal. Each process also improves its disk access latency by submittint its prefetches in batches so that the requests can be scheduled to optimize disk access performance. Our measurements show that this combination of techniques greatly improves the performance of the file system. We measured that the running time is reduced by 3% to 49% (average 26%) for single-process workloads and by 5% to 76% (average 32%) for multiprocess workloads.

The automatic improvement of locality in storage systems Disk I/O is increasingly the performance bottleneck in computer systems despite rapidly increasing disk data transfer rates. In this article, we propose Automatic Locality-Improving Storage (ALIS), an introspective storage system that automatically reorganizes selected disk blocks based on the dynamic reference stream to increase effective storage performance. ALIS is based on the observations that sequential data fetch is far more efficient than random access, that improving seek distances produces only marginal performance improvements, and that the increasingly powerful processors and large memories in storage systems have ample capacity to reorganize the data layout and redirect the accesses so as to take advantage of rapid sequential data transfer. Using trace-driven simulation with a large set of real workloads, we demonstrate that ALIS considerably outperforms prior techniques, improving the average read performance by up to 50&percnt; for server workloads and by about 15&percnt; for personal computer workloads. We also show that the performance improvement persists as disk technology evolves. Since disk performance in practice is increasing by only about 8&percnt; per year, the benefit of ALIS may correspond to as much as several years of technological progress.

Read Optimized File System Designs: A Performance Evaluation This paper presents a performance comparison of several file system allocation policies. The file systems are designed to provide high bandwidth between disks and main memory by taking advantage of parallelism in an underlying disk array, catering to large units of transfer, and minimizing the bandwidth dedicated to the transfer of meta data. All of the file systems described use a mul- tiblock allocation strategy which allows both large and small files to be allocated efficiently. Simulation results show that these multiblock policies result in systems that are able to utilize a large percentage of the underlying disk bandwidth; more than 90% in sequential cases. As general purpose systems are called upon to support more data intensive applications such as databases and super- computing, these policies offer an opportunity to provide superior performance to a larger class of users.

AirCloud: a cloud-based air-quality monitoring system for everyone We present the design, implementation, and evaluation of AirCloud -- a novel client-cloud system for pervasive and personal air-quality monitoring at low cost. At the frontend, we create two types of Internet-connected particulate matter (PM2:5) monitors -- AQM and miniAQM, with carefully designed mechanical structures for optimal air-flow. On the cloud-side, we create an air-quality analytics engine that learn and create models of air-quality based on a fusion of sensor data. This engine is used to calibrate AQMs and mini-AQMs in real-time, and infer PM2:5 concentrations. We evaluate AirCloud using 5 months of data and 2 month of continuous deployment, and show that AirCloud is able to achieve good accuracies at much lower cost than previous solutions. We also show three real applications built on top of AirCloud by 3rd party developers to further demonstrate the value of our system.

A Hybrid Approach for Speech Enhancement Using MoG Model and Neural Network Phoneme Classifier In this paper, we present a single-microphone speech enhancement algorithm. A hybrid approach is proposed merging the generative mixture of Gaussians MoG model and the discriminative deep neural network DNN. The proposed algorithm is executed in two phases, the training phase, which does not recur, and the test phase. First, the noise-free speech log-power spectral density is modeled as an MoG, representing the phoneme-based diversity in the speech signal. A DNN is then trained with phoneme labeled database of clean speech signals for phoneme classification with mel-frequency cepstral coefficients as the input features. In the test phase, a noisy utterance of an untrained speech is processed. Given the phoneme classification results of the noisy speech utterance, a speech presence probability SPP is obtained using both the generative and discriminative models. SPP-controlled attenuation is then applied to the noisy speech while simultaneously, the noise estimate is updated. The discriminative DNN maintains the continuity of the speech and the generative phoneme-based MoG preserves the speech spectral structure. Extensive experimental study using real speech and noise signals is provided. We also compare the proposed algorithm with alternative speech enhancement algorithms. We show that we obtain a significant improvement over previous methods in terms of speech quality measures. Finally, we analyze the contribution of all components of the proposed algorithm indicating their combined importance.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Capturing and Matching Dynamic Behaviour in Case-Based Reasoning In the telecommunications domain, reuse of service specifications is a major issue. However, it has proved difficult to modularise services because of the high degree of interaction between them. Direct application of formal logics to the specification of services has proved impractical because of the size of the services. However, much of this complexity stems from the details of implementatio n of the services; by contrast, the principal behaviours of a service are often approximated by simple varieties of logic which are easily accessible to users. We address the problem of determining, from a library of services, those which might be appropriate for reuse in constructing a new service. Simple behavioural sequences are used to provide features within a CBR system which matches these to behavioural examples supplied by users. By side-stepping the problem of formally specifying the entire service, we aim to promote greater reuse of services while avoiding a commitment to full logical specification. Non-mathematicians often have difficulty in expressing requirements formally. By using a CBR approach the user can sketch out simple, familiar behaviours and with these examples the system is able to retrieve relevant cases and interactively produce a formal requirements sketch capturing the new required behaviour. A case in the case library encapsulates a particular formalised behaviour in a simple logic which is sufficient to capture the key dynamic behaviours of the domain. With a simulator the user can evaluate the behaviour without being confronted with the formal representation itself. Our domain is telephone features such as call waiting, redirect call, call back. These telephone services are stored in the case library as cases, each consisting of a set of transition rules. In previous papers we have described the general architecture of the system (see for example (Funk & Robertson 1994)). In this paper we focus on matching dynamic behaviour and the formal representation of the cases.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Automatic Identification of Instrument Classes in Polyphonic and Poly-Instrument Audio.

Audio-based Music Classification with a Pretrained Convolutional Network.

Moving Beyond Feature Design: Deep Architectures and Automatic Feature Learning in Music Informatics.

Backpropagation Applied to Handwritten Zip Code Recognition. The ability of learning networks to generalize can be greatly enhanced by providing constraints from the task domain. This paper demonstrates how such constraints can be integrated into a backpropagation network through the architecture of the network. This approach has been successfully applied to the recognition of handwritten zip code digits provided by the U.S. Postal Service. A single network learns the entire recognition operation, going from the normalized image of the character to the final classification.

Convex Neural Networks Convexity has recently received a lot of attention in the machine learning community, and the lack of convexity has been seen as a major disad- vantage of many learning algorithms, such as multi-layer artificial neural networks. We show that training multi-layer neural networks in which the number of hidden units is learned can be viewed as a convex optimization problem. This problem involves an infinite number of variables, but can be solved by incrementally inserting a hidden unit at a time, each time finding a linear classifier that minimizes a weighted sum of errors.

Modeling Temporal Dependencies in High-Dimensional Sequences: Application to Polyphonic Music Generation and Transcription. We investigate the problem of modeling symbolic sequences of polyphonic music in a completely general piano-roll representation. We introduce a probabilistic model based on distribution estimators conditioned on a recurrent neural network that is able to discover temporal dependencies in high-dimensional sequences. Our approach outperforms many traditional models of polyphonic music on a variety of realistic datasets. We show how our musical language model can serve as a symbolic prior to improve the accuracy of polyphonic transcription.

Learning long-term dependencies with gradient descent is difficult Recurrent neural networks can be used to map input sequences to output sequences, such as for recognition, production or prediction problems. However, practical difficulties have been reported in training recurrent neural networks to perform tasks in which the temporal contingencies present in the input/output sequences span long intervals. We show why gradient based learning algorithms face an increasingly difficult problem as the duration of the dependencies to be captured increases. These results expose a trade-off between efficient learning by gradient descent and latching on information for long periods. Based on an understanding of this problem, alternatives to standard gradient descent are considered.

Deep Boltzmann Machines We present a new learning algorithm for Boltz- mann machines that contain many layers of hid- den variables. Data-dependent expectations are estimated using a variational approximation that tends to focus on a single mode, and data- independent expectations are approximated us- ing persistent Markov chains. The use of two quite different techniques for estimating the two types of expectation that enter into the gradient of the log-likelihood makes it practical to learn Boltzmann machines with multiple hidden lay- ers and millions of parameters. The learning can be made more efficient by using a layer-by-layer "pre-training" phase that allows variational in- ference to be initialized with a single bottom- up pass. We present results on the MNIST and NORB datasets showing that deep Boltzmann machines learn good generative models and per- form well on handwritten digit and visual object recognition tasks.

The effects of adding noise during backpropagation training on a generalization performance We study the effects of adding noise to the inputs, outputs, weight connections, and weight changes of multilayer feedforward neural networks during backpropagation training. We rigorously derive and analyze the objective functions that are minimized by the noise-affected training processes. We show that input noise and weight noise encourage the neural-network output to be a smooth function of the input or its weights, respectively. In the weak-noise limit, noise added to the output of the neural networks only changes the objective function by a constant. Hence, it cannot improve generalization. Input noise introduces penalty terms in the objective function that are related to, but distinct from, those found in the regularization approaches. Simulations have been performed on a regression and a classification problem to further substantiate our analysis. Input noise is found to be effective in improving the generalization performance for both problems. However, weight noise is found to be effective in improving the generalization performance only for the classification problem. Other forms of noise have practically no effect on generalization.

How to Construct Deep Recurrent Neural Networks. In this paper, we explore different ways to extend a recurrent neural network (RNN) to a \textit{deep} RNN. We start by arguing that the concept of depth in an RNN is not as clear as it is in feedforward neural networks. By carefully analyzing and understanding the architecture of an RNN, however, we find three points of an RNN which may be made deeper; (1) input-to-hidden function, (2) hidden-to-hidden transition and (3) hidden-to-output function. Based on this observation, we propose two novel architectures of a deep RNN which are orthogonal to an earlier attempt of stacking multiple recurrent layers to build a deep RNN (Schmidhuber, 1992; El Hihi and Bengio, 1996). We provide an alternative interpretation of these deep RNNs using a novel framework based on neural operators. The proposed deep RNNs are empirically evaluated on the tasks of polyphonic music prediction and language modeling. The experimental result supports our claim that the proposed deep RNNs benefit from the depth and outperform the conventional, shallow RNNs.

The Multi-Queue Replacement Algorithm for Second Level Buffer Caches This paper reports our research results that improve second level buffer cache performance. Several previous studies have shown that a good single level cache replacement algorithm such as LRU does not work well with second level buffer caches. Second level buffer caches have different access pattern from first level buffer caches because Accesses to second level buffer caches are actually misses from first level buffer caches.The paper presents our study of second level buffer cache access patterns using four large traces from various servers. We also introduce a new second level buffer cache replacement algorithm called Multi-Queue (MQ). Our trace-driven simulation results show that MQ performs better than all seven tested alternatives. Our implementation on a real storage system validates these results.

Fast consistency checking for the Solaris file system Our Netra NFS group at Sun set out to solve the challenging problem of providing remote Network File System (NFS) service with high performance and availability. An NFS server must guarantee the permanence of changes to the file system before acknowledging an NFS request. Thus, the server's underlying local file system must perform update operations synchronously to stable storage with potentially high latency. Our solution to this problem involves using the Solaris Unix File System (UFS), derived from the Berkeley Fast File System (FFS), in conjunction with nonvolatile RAM (NVRAM) as fast stable storage. We evaluated the system using the LADDIS benchmark and as a result, developed a cacheing technique for block-mapping information that gave us a 23% increase in measured server throughput in our standard RAID-5 server configuration. With recent increases in disk capacity and RAID technology, filesystem sizes have reached a point not imagined by the FFS designers, requiring an approach to checking file-system consistency that does not grow proportionately with file-system size. We examined several log-based solutions to providing fast crash recovery, but none could use the NVRAM effectively and meet our performance requirements. As an alternative, we developed an approach that uses UFS but maintains file-system working-set information, so that the consistency checker needs to examine only the active portions of a file system. This approach met our performance goals and also reduced file-system consistency-checking times to between 3% and 25% of those in the original UFS implementation.

Long-distance mutual exclusion for propositional planning The use of mutual exclusion (mutex) has led to significant advances in propositional planning. However, previous mutex can only detect pairs of actions or facts that cannot be arranged at the same time step. In this paper, we introduce a new class of constraints that significantly generalizes mutex and can be efficiently computed. The proposed long-distance mutual exclusion (londex) can capture constraints over actions and facts not only at the same time step but also across multiple steps. Londex provides a powerful and general approach for improving planning efficiency. As an application, we have integrated londex into SATPLAN04, a leading optimal planner. Experimental results show that londex can effectively prune the search space and reduce the planning time. The resulting planner, MaxPlan, has won the First Place Award in the Optimal Track of the 5th International Planning Competition.

Incremental learning by message passing in hierarchical temporal memory Hierarchical temporal memory HTM is a biologically inspired framework that can be used to learn invariant representations of patterns in a wide range of applications. Classical HTM learning is mainly unsupervised, and once training is completed, the network structure is frozen, thus making further training i.e., incremental learning quite critical. In this letter, we develop a novel technique for HTM incremental supervised learning based on gradient descent error minimization. We prove that error backpropagation can be naturally and elegantly implemented through native HTM message passing based on belief propagation. Our experimental results demonstrate that a two-stage training approach composed of unsupervised pretraining and supervised refinement is very effective both accurate and efficient. This is in line with recent findings on other deep architectures.

Challenges designing next-generation middleware systems This framework promises new classes of service, especially in terms of security, for policy-based development of distributed and collaborative applications.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

Concurrent actions in the situation calculus We propose a representation of Concurrent actions; rather than invent a new formalism, we model them within the standard situation calculus by introducing the notions of global actions and primitive actions, whose relationship is analogous to' that between situations and fluents. The result is a framework in which situations and actions play quite symmetric roles. The rich structure of actions gives rise to' a new problem, which, due to' this symmetry between actions and situations, is analogous to' the traditional frame problem. In [Lin and Shoham 1991] we provided a solution to' the frame problem based on a formal adequacy criterion called "epistemological completeness." Here we show how to' solve the new problem based on the same adequacy criterion.

Exploring Gate-Limited Analytical Models for High Performance Network Storage Servers

Parameterized complexity for the database theorist

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Optimizing Data Intensive GPGPU Computations for DNA Sequence Alignment. MUMmerGPU uses highly-parallel commodity graphics processing units (GPU) to accelerate the data-intensive computation of aligning next generation DNA sequence data to a reference sequence for use in diverse applications such as disease genotyping and personal genomics. MUMmerGPU 2.0 features a new stackless depth-first-search print kernel and is 13× faster than the serial CPU version of the alignment code and nearly 4× faster in total computation time than MUMmerGPU 1.0. We exhaustively examined 128 GPU data layout configurations to improve register footprint and running time and conclude higher occupancy has greater impact than reduced latency. MUMmerGPU is available open-source at http://mummergpu.sourceforge.net.

Speeding up subset seed algorithm for intensive protein sequence comparison Abstract—Sequence similarity search is a common and re- peated task in molecular biology. The rapid growth of genomic databases leads to the need of speeding up the treatment of this task. In this paper, we present a subset seed algorithm for intensive protein sequence comparison. We have accelerated this algorithm by using indexing technique and fine grained parallelism of GPU and SIMD instructions. We have implemented two programs: iBLASTP, iTBLASTN. The GPU (SIMD) imple- mentation of the two programs achieves a speed up ranging from 5.5 to 10 (4 to 5.6) compared to the BLASTP and TBLASTN of the BLAST program family, with comparable sensitivity.

The Astral Compendium For Protein Structure And Sequence Analysis The ASTRAL compendium provides several databases and tools to aid in the analysis of protein structures, particularly through the use of their sequences. The SPACI scores included in the system summarize the overall characteristics of a protein structure. A structural alignments database indicates residue equivalencies in superimposed protein domain structures, The PDB sequence-map files provide a linkage between the amino acid sequence of the molecule studied (SEQRES records in a database entry) and the sequence of the atoms experimentally observed in the structure (ATOM records). These maps are combined with information in the SCOP database to provide sequences of protein domains. Selected subsets of the domain database, with varying degrees of similarity measured in several different ways, are also available. ASTRAL may be accessed at http://astral.stanford.edu/.

Accelerating BLASTP on the Cell Broadband Engine The enormous growth of biological sequence databases has caused bioinformatics to be rapidly moving towards a data-intensive, computational science. As a result, the computational power needed by bioinformatics applications is growing rapidly as well. The recent emergence of low cost parallel accelerator technologies has made it possible to reduce execution times of many bioinformatics applications. In this paper, we demonstrate how the PlayStation®3, powered by the Cell Broadband Engine, can be used as an efficient computational platform to accelerate the popular BLASTP algorithm.

Efficient parallel algorithm for multiple sequence alignments with regular expression constraints on graphics processing units Multiple sequence alignments with constraints has become an important problem in computational biology. The concept of constrained sequence alignment is proposed to incorporate the biologist's domain knowledge into sequence alignments such that the user-specified residues/segments are aligned together in the alignment results. Over the past decade, a series of constrained multiple sequence alignment tools were proposed in the literature. RE-MuSiC is the newest tool with the regular expression constraints and useful for a wide range of biological applications. However, the computation time of REMuSiC is large for a large amount of sequences or long sequences and this problem limits the application usage. Therefore, in this paper, a tool, GPU-REMuSiC v1.0, is proposed to reduce the computation time of RE-MuSiC by using the graphics processing units with CUDA. GPU-REMuSiC v1.0 can achieve 29× speedups for overall computation time by the experimental results.

Mapping of BLASTP Algorithm onto GPU Clusters Searching protein sequence database is a fundamental and often repeated task in computational biology and bioinformatics. However, the high computational cost and long runtime of many database scanning algorithms on sequential architectures heavily restrict their applications for large-scale protein databases, such as GenBank. The continuing exponential growth of sequence databases and the high rate of newly generated queries further deteriorate the situation and establish a strong requirement for time-efficient scalable database searching algorithms. In this paper, we demonstrate how GPU clusters, powered by the Compute Unified Device Architecture (CUDA), OpenMP, and MPI parallel programming models can be used as an efficient computational platform to accelerate the popular BLASTP algorithm. Compared to GPU-BLAST 1.0-2.2.24, our implementation achieves speedups up to 1.6 on a single GPU and up to 6.6 on the 6 GPUs of a Tesla S1060 quad-GPU computing system. The source code is available at: http://sites.google.com/site/liuweiguohome/mpicuda-blastp

Proceedings of the 24th International Conference on Supercomputing, 2010, Tsukuba, Ibaraki, Japan, June 2-4, 2010

Reconfigurable computing: a survey of systems and software Due to its potential to greatly accelerate a wide variety of applications, reconfigurable computing has become a subject of a great deal of research. Its key feature is the ability to perform computations in hardware to increase performance, while retaining much of the flexibility of a software solution. In this survey, we explore the hardware aspects of reconfigurable computing machines, from single chip architectures to multi-chip systems, including internal structures and external coupling. We also focus on the software that targets these machines, such as compilation tools that map high-level algorithms directly to the reconfigurable substrate. Finally, we consider the issues involved in run-time reconfigurable systems, which reuse the configurable hardware during program execution.

Histograms of Oriented Gradients for Human Detection We study the question of feature sets for robust visual object recognition, adopting linear SVM based human detection as a test case. After reviewing existing edge and gradient based descriptors, we show experimentally that grids of Histograms of Oriented Gradient (HOG) descriptors significantly outperform existing feature sets for human detection. We study the influence of each stage of the computation on performance, concluding that fine-scale gradients, fine orientation binning, relatively coarse spatial binning, and high-quality local contrast normalization in overlapping descriptor blocks are all important for good results. The new approach gives near-perfect separation on the original MIT pedestrian database, so we introduce a more challenging dataset containing over 1800 annotated human images with a large range of pose variations and backgrounds.

A unified architecture for natural language processing: deep neural networks with multitask learning We describe a single convolutional neural network architecture that, given a sentence, outputs a host of language processing predictions: part-of-speech tags, chunks, named entity tags, semantic roles, semantically similar words and the likelihood that the sentence makes sense (grammatically and semantically) using a language model. The entire network is trained jointly on all these tasks using weight-sharing, an instance of multitask learning. All the tasks use labeled data except the language model which is learnt from unlabeled text and represents a novel form of semi-supervised learning for the shared tasks. We show how both multitask learning and semi-supervised learning improve the generalization of the shared tasks, resulting in state-of-the-art-performance.

Distributed, object-based programming systems The development of distributed operating systems and object-based programming languages makes possible an environment in which programs consisting of a set of interacting modules, or objects, may execute concurrently on a collection of loosely coupled processors. An object-based programming language encourages a methodology for designing and creating a program as a set of autonomous components, whereas a distributed operating system permits a collection of workstations or personal computers to be treated as a single entity. The amalgamation of these two concepts has resulted in systems that shall be referred to as distributed, object-based programming systems. This paper discusses issues in the design and implementation of such systems. Following the presentation of fundamental concepts and various object models, issues in object management, object interaction management, and physical resource management are discussed. Extensive examples are drawn from existing systems.

Complexity of Data Tree Patterns over XML Documents We consider Boolean combinations of data tree patterns as a specification and query language for XML documents. Data tree patterns are tree patterns plus variable (in)equalities which express joins between attribute values. Data tree patterns are a simple and natural formalism for expressing properties of XML documents. We consider first the model checking problem (query evaluation), we show that it is DP-complete in general and already NP-complete when we consider a single pattern. We then consider the satisfiability problem in the presence of a DTD. We show that it is in general undecidable and we identify several decidable fragments.

A Markov Decision Problem Approach to Goal Attainment A new Markov decision problem (MDP)-based method for managing goal attainment (GA), which is the process of planning and controlling actions that are related to the achievement of a set of defined goals in the presence of resource and time constraints, is proposed. Specifically, we address the problem as one of optimally selecting a sequence of actions to transform the system and/or its environment from an initial state to a desired state. We begin with a method of explicitly mapping an action-GA graph to an MDP graph and developing a dynamic programming (DP) recursion to solve the MDP problem. For larger problems having exponential complexity with respect to the number of goals, we propose guided search algorithms such as AO*, AOepsiv*, and greedy search techniques, whose search power rests on the efficiency of their heuristic evaluation functions (HEFs). Our contribution in this part stems from the introduction of a new problem-specific HEF to aid the search process. We demonstrate reductions in the computational costs of the proposed techniques through performance comparison with standard DP techniques. We conclude this paper with a method to address situations in which alternative strategies (e.g., second best) are required. The new extended AO* algorithm identifies alternative control sequences for attaining the organizational goals.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

A cost effective fault-tolerant scheme for RAIDs The rapid progress in mass storage technology has made it possible for designers to implement large data storage systems for a variety of applications. One of the efficient ways to build large storage systems is to use RAIDs as basic storage modules. In general, the data can be recovered in RAIDs only when one error occurs. But in large RAIDs systems, the fault probability will increase when the number of disks increases, and the use of disks with big storage capacity will cause the recovering time to prolong, thus the probability of the second disk's fault will increase. Therefore, it is necessary to develop methods to recover data when two or more errors have occurred. In this paper, a fault tolerant scheme is proposed based on extended Reed-Solomon code, a recovery procedure is designed to correct up to two errors which is implemented by software and hardware together, and the scheme is verified by computer simulation. In this scheme, only two redundant disks are used to recover up to two disks' fault. The encoding and decoding methods, and the implementation based on software and hardware are described. The application of the scheme in software RAIDs that are built in cluster computers are also described. Compared with the existing methods such as EVENODD and DH, the proposed scheme has distinct improvement in implementation and redundancy.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Prediction model of hot metal temperature for blast furnace based on improved multi-layer extreme learning machine In the blast furnace production site, the disposable thermocouple is used to measure the hot metal temperature. However, this method is not only inconvenient for continuous data acquisition but also costly for the use of one-time thermocouple. Hence, this paper establishes a prediction model to predict the hot metal temperature. Before the prediction model is established, the corresponding factors of influencing the hot metal temperature are selected, and the noises of production data are removed. In this paper, multi-layer extreme learning machine (ML-ELM) is used as the prediction algorithm of the prediction model. However, the input weights, hidden layer weights and hidden biases of ML-ELM are randomly selected, and the solution of the output weights is based on them, which makes ML-ELM inevitably have a set of non-optimal or unnecessary weights and biases. In addition, ML-ELM may suffer from over-fitting problem. Hence, this paper uses the adaptive particle swarm optimization (APSO) and the ensemble model to improve ML-ELM, and the improved algorithm is named as EAPSO-ML-ELM. APSO can optimize the selections of the input weights, hidden layer weights and hidden biases, the ensemble model can alleviate the over-fitting problem, i.e., this paper combines several of the optimized ML-ELMs which have different input weights, hidden layer weights and hidden biases. Finally, this paper also uses other algorithms to establish the prediction model, and simulation results demonstrate that the prediction model based on EAPSO-ML-ELM has better prediction accuracy and generalization performance.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Transformation Pursuit for Image Classification A simple approach to learning invariances in image clas- sification consists in augmenting the training set with transformed versions of the original images. However, given a large set of possible transformations, selecting a com- pact subset is challenging. Indeed, all transformations are not equally informative and adding uninformative transfor- mations increases training time with no gain in accuracy. We propose a principled algorithm--Image Transformation Pursuit (ITP)--for the automatic selection of a compact set of transformations. ITP works in a greedy fashion, by se- lecting at each iteration the one that yields the highest accuracy gain. ITP also allows to efficiently explore complex transformations, that combine basic transformations. We report results on two public benchmarks: the CUB dataset of bird images and the ImageNet 2010 challenge. Using Fisher Vector representations, we achieve an improvement from 28.2% to 45.2% in top-1 accuracy on CUB, and an im- provement from 70.1% to 74.9% in top-5 accuracy on Im- ageNet. We also show significant improvements for deep convnet features: from 47.3% to 55.4% on CUB and from 77.9% to 81.4% on ImageNet.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Accurate Deep Neural Network Model To Detect Cardiac Arrhythmia On More Than 10,0 00 Individual Subject Ecg Records Background and objective: Cardiac arrhythmia, which is an abnormal heart rhythm, is a common clinical problem in cardiology. Detection of arrhythmia on an extended duration electrocardiogram (ECG) is done based on initial algorithmic software screening, with final visual validation by cardiologists. It is a time consuming and subjective process. Therefore, fully automated computer-assisted detection systems with a high degree of accuracy have an essential role in this task. In this study, we proposed an effective deep neural network (DNN) model to detect different rhythm classes from a new ECG database.Methods: Our DNN model was designed for high performance on all ECG leads. The proposed model, which included both representation learning and sequence learning tasks, showed promising results on all 12-lead inputs. Convolutional layers and sub-sampling layers were used in the representation learning phase. The sequence learning part involved a long short-term memory (LSTM) unit after representation of learning layers.Results: We performed two different class scenarios, including reduced rhythms (seven rhythm types) and merged rhythms (four rhythm types) according to the records from the database. Our trained DNN model achieved 92.24% and 96.13% accuracies for the reduced and merged rhythm classes, respectively.Conclusion: Recently, deep learning algorithms have been found to be useful because of their high performance. The main challenge is the scarcity of appropriate training and testing resources because model performance is dependent on the quality and quantity of case samples. In this study, we used a new public arrhythmia database comprising more than 10,000 records. We constructed an efficient DNN model for automated detection of arrhythmia using these records. (C) 2020 Elsevier B.V. All rights reserved.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

A new intra-disk redundancy scheme for high-reliability RAID storage systems in the presence of unrecoverable errors Today's data storage systems are increasingly adopting low-cost disk drives that have higher capacity but lower reliability, leading to more frequent rebuilds and to a higher risk of unrecoverable media errors. We propose an efficient intradisk redundancy scheme to enhance the reliability of RAID systems. This scheme introduces an additional level of redundancy inside each disk, on top of the RAID redundancy across multiple disks. The RAID parity provides protection against disk failures, whereas the proposed scheme aims to protect against media-related unrecoverable errors. In particular, we consider an intradisk redundancy architecture that is based on an interleaved parity-check coding scheme, which incurs only negligible I/O performance degradation. A comparison between this coding scheme and schemes based on traditional Reed--Solomon codes and single-parity-check codes is conducted by analytical means. A new model is developed to capture the effect of correlated unrecoverable sector errors. The probability of an unrecoverable failure associated with these schemes is derived for the new correlated model, as well as for the simpler independent error model. We also derive closed-form expressions for the mean time to data loss of RAID-5 and RAID-6 systems in the presence of unrecoverable errors and disk failures. We then combine these results to characterize the reliability of RAID systems that incorporate the intradisk redundancy scheme. Our results show that in the practical case of correlated errors, the interleaved parity-check scheme provides the same reliability as the optimum, albeit more complex, Reed--Solomon coding scheme. Finally, the I/O and throughput performances are evaluated by means of analysis and event-driven simulation.

RAID triple parity RAID triple parity (RTP) is a new algorithm for protecting against three-disk failures. It is an extension of the double failure correction Row-Diagonal Parity code. For any number of data disks, RTP uses only three parity disks. This is optimal with respect to the amount of redundant information required and accessed. RTP uses XOR operations and stores all data un-encoded. The algorithm's parity computation complexity is provably optimal. The decoding complexity is also much lower than that of existing comparable codes. This paper also describes a symmetric variant of the algorithm where parity computation is identical to triple reconstruction.

Beyond MTTDL: A Closed-Form RAID 6 Reliability Equation We introduce a new closed-form equation for estimating the number of data-loss events for a redundant array of inexpensive disks in a RAID-6 configuration. The equation expresses operational failures, their restorations, latent (sector) defects, and disk media scrubbing by time-based distributions that can represent non-homogeneous Poisson processes. It uses two-parameter Weibull distributions that allows the distributions to take on many different shapes, modeling increasing, decreasing, or constant occurrence rates. This article focuses on the statistical basis of the equation. It also presents time-based distributions of the four processes based on an extensive analysis of field data collected over several years from 10,000s of commercially available systems with 100,000s of disk drives. Our results for RAID-6 groups of size 16 indicate that the closed-form expression yields much more accurate results compared to the MTTDL reliability equation and matching computationally-intensive Monte Carlo simulations.

On the impact of disk scrubbing on energy savings The increasing use of computers for saving valuable data imposes stringent reliability constraints on storage systems. Reliability improvement via use of redundancy is a common practice. As the disk capacity improves, advanced techniques such as disk scrubbing are being employed to proactively fix latent sector errors. These techniques utilize the disk idle time for reliability improvement. However, the idle time is a key to dynamic energy management that detects such idle periods and turns-off the disks to save energy. In this paper, we are concerned with the distribution of the disk idle periods between reliability and energy management tasks. For this purpose, we define a new metric, energy-reliability product (ERP), to capture the effect of one technique on the other. Our initial investigation using trace-driven simulations of typical enterprise applications shows that the ERP is a suitable metric for identifying efficient idle period utilization. Thus, ERP can facilitate development of systems that provide both reliability and energy managements.

Disk Scrubbing Versus Intradisk Redundancy for RAID Storage Systems Two schemes proposed to cope with unrecoverable or latent media errors and enhance the reliability of RAID systems are examined. The first scheme is the established, widely used, disk scrubbing scheme, which operates by periodically accessing disk drives to detect media-related unrecoverable errors. These errors are subsequently corrected by rebuilding the sectors affected. The second scheme is the recently proposed intradisk redundancy scheme, which uses a further level of redundancy inside each disk, in addition to the RAID redundancy across multiple disks. A new model is developed to evaluate the extent to which disk scrubbing reduces the unrecoverable sector errors. The probability of encountering unrecoverable sector errors is derived analytically under very general conditions regarding the characteristics of the read/write process of uniformly distributed random workloads and for a broad spectrum of disk scrubbing schemes, which includes the deterministic and random scrubbing schemes. We show that the deterministic scrubbing scheme is the most efficient one. We also derive closed-form expressions for the percentage of unrecoverable sector errors that the scrubbing scheme detects and corrects, the throughput performance, and the minimum scrubbing period achievable under operation with random, uniformly distributed I/O requests. Our results demonstrate that the reliability improvement due to disk scrubbing depends on the scrubbing frequency and the load of the system, and, for heavy-write workloads, may not reach the reliability level achieved by a simple interleaved parity-check (IPC)-based intradisk redundancy scheme, which is insensitive to the load. In fact, for small unrecoverable sector error probabilities, the IPC-based intradisk redundancy scheme achieves essentially the same reliability as that of a system operating without unrecoverable sector errors. For heavy loads, the reliability achieved by the scrubbing scheme can be orders of magnitude less than that of the intradisk redundancy scheme. Finally, the I/O and throughput performances are evaluated by means of analysis and event-driven simulation.

Multi-level RAID for very large disk arrays Very Large Disk Arrays - VLDAs have been developed to cope with the rapid increase in the volume of data generated requiring ultrareliable storage. Bricks or Storage Nodes - SNs holding a dozen or more disks are cost effective VLDA building blocks, since they cost less than traditional disk arrays. We utilize the Multilevel RAID - MRAID paradigm for protecting both SNs and their disks. Each SN is a k-disk-failure-tolerant kDFT array, while replication or l-node failure tolerance - lNFTs paradigm is applied at the SN level. For example, RAID1(M)/5(N) denotes a RAID1 at the higher level with a degree of replication M and each virtual disk is an SN configured as a RAID5 with N physical disks. We provide the data layout for RAID5/5 and RAID6/5 MRAIDs and give examples of updating data and recovering lost data. The former requires storage transactions to ensure the atomicity of storage updates. We discuss some weaknesses in reliability modeling in RAID5 and give examples of an asymptotic expansion method to compare the reliability of several MRAID organizations. We outline the reliability analysis of Markov chain models of VLDAs and briefly report on conclusions from simulation results. In Conclusions we outline areas for further research.

Reliability analysis of deduplicated and erasure-coded storage Space efficiency and data reliability are two primary concerns for modern storage systems. Chunk-based deduplication, which breaks up data objects into single-instance chunks that can be shared across objects, is an effective method for saving storage space. However, deduplication affects data reliability because an object's constituent chunks are often spread across a large number of disks, potentially decreasing the object's reliability. Therefore, an important problem in deduplicated storage is how to achieve space efficiency yet maintain each object's original reliability. In this paper, we present initial results on the reliability analysis of HP-KVS, a deduplicated key-value store that allows each object to specify its own reliability level and that uses software erasure coding for data reliability. The combination of deduplication and erasure coding gives rise to several interesting research problems. We show how to compare the reliability of erasure codes with different parameters and how to analyze the reliability of a big data object given its constituent parts' reliabilities. We also present a method for system designers to determine under what conditions deduplication will save space for erasure-coded data.

EVENODD: an optimal scheme for tolerating double disk failures in RAID architectures We present a novel method, that we call EVENODD, for tolerating up to two disk failures in RAID architectures. EVENODD is the first known scheme for tolerating double disk failures that is optimal with regard to both storage and performance. EVENODD employs the addition of only two redundant disks and consists of simple exclusive-OR computations. A major advantage of EVENODD is that it only requires parity hardware, which is typically present in standard RAID-5 controllers. Hence, EVENODD can be implemented on standard RAID-5 controllers without any hardware changes. The only previously known scheme that employes optimal redundant storage (i.e. two extra disks) is based on Reed-Solomon (RS) error-correcting codes, requires computation over finite fields and results in a more complex implementation. For example, we show that the number of exclusive-OR operations involved in implementing EVENODD in a disk array with 15 disks is about 50% of the number required when using the RS scheme.

An optimality proof of the LRU-K page replacement algorithm This paper analyzes a recently published algorithm for page replacement in hierarchical paged memory systems [O'Neil et al. 1993]. The algorithm is called the LRU-K method, and reduces to the well-known LRU (Least Recently Used) method for K = 1. Previous work [O'Neil et al. 1993; Weikum et al. 1994; Johnson and Shasha 1994] has shown the effectiveness for K > 1 by simulation, especially in the most common case of K = 2. The basic idea in LRU-K is to keep track of the times of the last K references to memory pages, and to use this statistical information to rank-order the pages as to their expected future behavior. Based on this the page replacement policy decision is made: which memory-resident page to replace when a newly accessed page must be read into memory. In the current paper, we prove, under the assumptions of the independent reference model, that LRU-K is optimal. Specifically we show: given the times of the (up to) K most recent references to each disk page, no other algorithm A making decisions to keep pages in a memory buffer holding n - 1 pages based on this infomation can improve on the expected number of I/Os to access pages over the LRU-K algorithm using a memory buffer holding n pages. The proof uses the Bayesian formula to relate the space of actual page probabilities of the model to the space of observable page numbers on which the replacement decision is acutally made.

The logical disk: a new approach to improving file systems The Logical Disk (LD) defines a new interface to disk storage that separates file management and disk management by using logical block numbers and block lists. The LD interface is designed to support multiple file systems and to allow multiple implementations, both of which are important given the increasing use of kernels that support multiple operating system personalities.A log-structured implementation of LD (LLD) demonstrates that LD can be implemented efficiently. LLD adds about 5% to 10% to the purchase cost of a disk for the main memory it requires. Combining LLD with an existing file system results in a log-structured file system that exhibits the same performance characteristics as the Sprite log-structured file system.

Human-level control through deep reinforcement learning. The theory of reinforcement learning provides a normative account', deeply rooted in psychological' and neuroscientifie perspectives on animal behaviour, of how agents may optimize their control of an environment. To use reinforcement learning successfully in situations approaching real-world complexity, however, agents are confronted with a difficult task: they must derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations. Remarkably, humans and other animals seem to solve this problem through a harmonious combination of reinforcement learning and hierarchical sensory processing systems4'5, the former evidenced by a wealth of neural data revealing notable parallels between the phasic signals emitted by dopaminergic neurons and temporal difference reinforcement learning algorithms'. While reinforcement learning agents have achieved some successes in a variety of domains", their applicability has previously been limited to domains in which useful features can be handcrafted, or to domains with fully observed, low-dimensional state spaces. Here we use recent advances in training deep neural networks'" to develop a novel artificial agent, termed a deep Q-network, that can learn successful policies directly from high-dimensional sensory inputs using end-to-end reinforcement learning. We tested this agent on the challenging domain of classic Atari 2600 games". We demonstrate that the deep Q-network agent, receiving only the pixels and the game score as inputs, was able to surpass the performance of all previous algorithms and achieve a level comparable to that of a professional human games tester across a set of 49 games, using the same algorithm, network architecture and hyperparameters. This work bridges the divide between high-dimensional sensory inputs and actions, resulting in the first artificial agent that is capable of learning to excel at a diverse array of challenging tasks.

Deductive Plan Generation Abstract: of nonmonotonic inference rules to tackle the frame problem [19]. He uses adefault rule called law of inertia which states that a proposition does not changeits value when executing an action unless the contrary is known.Some years ago we developed a modified version of the connection methodto solve the frame problem without the need of any frame axioms [2]. In thelinear connection method proofs are restricted such that each literal is connectedat most once [2, 4, 3]. Thus, connecting a...

STEP: Sequentiality and Thrashing Detection Based Prefetching to Improve Performance of Networked Storage Servers State-of-the-art networked storage servers are equipped with increasingly powerful computing capability and large DRAMmemory as storage caches. However, their contribution to the performance improvement of networked storage system has become increasingly limited. This is because the client-side memory sizes are also increasing, which reduces capacity misses in the client buffer caches as well as access locality in the storage servers, thus weakening the caching effectiveness of server storage caches. Proactive caching in storage servers is highly desirable to reduce cold misses in clients. We propose an effective way to improve the utilization of storage server resources through prefetching in storage servers for clients. In particular, our design well utilizes two unique strengths of networked storage servers which are not leveraged in existing storage server prefetching schemes. First, powerful storage servers have idle CPU cycles, under-utilized disk bandwidth, and abundant memory space, providing many opportunities for aggressive disk data prefetching. Second, the servers have the knowledge about high-latency operations in storage devices, such as disk head positioning, which enables efficient disk data prefetching based on an accurate cost-benefit analysis of prefetch operations.We present STEP - a Sequentiality and Thrashing dEtection based Prefetching scheme, and its implementation with Linux Kernel 2.6.16. Our performance evaluation by replaying Storage Performance Council (SPC)'s OLTP traces shows that server performance improvements are up to 94% with an average of 25%. Improvements with frequently used Unix applications are up to 53% with an average of 12%. Our experiments also show that STEP has little effect on workloads with random access patterns, such as SPC' Web- Search traces.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Shallow vs. Deep Sum-Product Networks. We investigate the representational power of sum-product networks (computation networks analogous to neural networks, but whose individual units compute either products or weighted sums), through a theoretical analysis that compares deep (multiple hidden layers) vs. shallow (one hidden layer) architectures. We prove there exist families of functions that can be represented much more efficiently with a deep network than with a shallow one, i.e. with substantially fewer hidden units. Such results were not available until now, and contribute to motivate recent research involving learning of deep sum-product networks, and more generally motivate research in Deep Learning.

Does diversity improve deep learning? In this work, we carry out a first exploration of the possibility of increasing the performance of Deep Neural Networks (DNNs) by applying diversity techniques to them. Since DNNs arc usually very strong, weakening them can be important for this purpose. This paper includes experimental evidence of the effectiveness of binarizing multi-class problems to make beneficial the application of bagging to Denoising Auto-Encoding-Based DNNs for solving the classical MNIST problem. Many research opportunities appear following the diversification idea: We mention some of the most relevant lines at the end of this contribution.

Instant Learning: Parallel Deep Neural Networks and Convolutional Bootstrapping Although deep neural networks (DNN) are able to scale with direct advances in computational power (e.g., memory and processing speed), they are not well suited to exploit the recent trends for parallel architectures. In particular, gradient descent is a sequential process and the resulting serial dependencies mean that DNN training cannot be parallelized effectively. Here, we show that a DNN may be replicated over a massive parallel architecture and used to provide a cumulative sampling of local solution space which results in rapid and robust learning. We introduce a complimentary convolutional bootstrapping approach that enhances performance of the parallel architecture further. Our parallelized convolutional bootstrapping DNN out-performs an identical fully-trained traditional DNN after only a single iteration of training.

Learning a good representation with unsymmetrical auto-encoder Auto-encoders play a fundamental role in unsupervised feature learning and learning initial parameters of deep architectures for supervised tasks. For given input samples, robust features are used to generate robust representations from two perspectives: (1) invariant to small variation of samples and (2) reconstruction by decoders with minimal error. Traditional auto-encoders with different regularization terms have symmetrical numbers of encoder and decoder layers, and sometimes parameters. We investigate the relation between the number of layers and propose an unsymmetrical structure, i.e., an unsymmetrical auto-encoder (UAE), to learn more effective features. We present empirical results of feature learning using the UAE and state-of-the-art auto-encoders for classification tasks with a range of datasets. We also analyze the gradient vanishing problem mathematically and provide suggestions for the appropriate number of layers to use in UAEs with a logistic activation function. In our experiments, UAEs demonstrated superior performance with the same configuration compared to other auto-encoders.

Construction of neural networks for realization of localized deep learning. The subject of deep learning has recently attracted users of machine learning from various disciplines, including: medical diagnosis and bioinformatics, financial market analysis and online advertisement, speech and handwriting recognition, computer vision and natural language processing, time series forecasting, and search engines. However, theoretical development of deep learning is still at its infancy. The objective of this paper is to introduce a deep neural network (also called deep-net) approach to localized manifold learning, with each hidden layer endowed with a specific learning task. For the purpose of illustrations, we only focus on deep-nets with three hidden layers, with the first layer for dimensionality reduction, the second layer for bias reduction, and the third layer for variance reduction. A feedback component is also designed to deal with outliers. The main theoretical result in this paper is the order $mathcal Oleft(m^{-2s/(2s+d)}right)$ of approximation of the regression function with regularity $s$, in terms of the number $m$ of sample points, where the (unknown) manifold dimension $d$ replaces the dimension $D$ of the sampling (Euclidean) space for shallow nets.

Model complexities of shallow networks representing highly varying functions Model complexities of shallow (i.e., one-hidden-layer) networks representing highly varying multivariable { - 1 , 1 } -valued functions are studied in terms of variational norms tailored to dictionaries of network units. It is shown that bounds on these norms define classes of functions computable by networks with constrained numbers of hidden units and sizes of output weights. Estimates of probabilistic distributions of values of variational norms with respect to typical computational units, such as perceptrons and Gaussian kernel units, are derived via geometric characterization of variational norms combined with the probabilistic Chernoff Bound. It is shown that almost any randomly chosen { - 1 , 1 } -valued function on a sufficiently large d-dimensional domain has variation with respect to perceptrons depending on d exponentially.

Tensor Deep Stacking Networks A novel deep architecture, the tensor deep stacking network (T-DSN), is presented. The T-DSN consists of multiple, stacked blocks, where each block contains a bilinear mapping from two hidden layers to the output layer, using a weight tensor to incorporate higher order statistics of the hidden binary (½0; 1) features. A learning algorithm for the T-DSN’s weight matrices and tensors is developed and described in which the main parameter estimation burden is shifted to a convex subproblem with a closed-form solution. Using an efficient and scalable parallel implementation for CPU clusters, we train sets of T-DSNs in three popular tasks in increasing order of the data size: handwritten digit recognition using MNIST (60k), isolated state/phone classification and continuous phone recognition using TIMIT (1.1 m), and isolated phone classification using WSJ0 (5.2 m). Experimental results in all three tasks demonstrate the effectiveness of the T-DSN and the associated learning methods in a consistent manner. In particular, a sufficient depth of the T-DSN, a symmetry in the two hidden layers structure in each T-DSN block, our model parameter learning algorithm, and a softmax layer on top of T-DSN are shown to have all contributed to the low error rates observed in the experiments for all three tasks.

Deep learning for healthcare decision making with EMRs Computer aid technology is widely applied in decision-making and outcome assessment of healthcare delivery, in which modeling knowledge and expert experience is technically important. However, the conventional rule-based models are incapable of capturing the underlying knowledge because they are incapable of simulating the complexity of human brains and highly rely on feature representation of problem domains. Thus we attempt to apply a deep model to overcome this weakness. The deep model can simulate the thinking procedure of human and combine feature representation and learning in a unified model. A modified version of convolutional deep belief networks is used as an effective training method for large-scale data sets. Then it is tested by two instances: a dataset on hypertension retrieved from a HIS system, and a dataset on Chinese medical diagnosis and treatment prescription from a manual converted electronic medical record (EMR) database. The experimental results indicate that the proposed deep model is able to reveal previously unknown concepts and performs much better than the conventional shallow models.

Automatic Identification of Instrument Classes in Polyphonic and Poly-Instrument Audio.

Principled Hybrids of Generative and Discriminative Models When labelled training data is plentiful, discriminative techniques are widely used since they give excellent generalization performance. However, for large-scale applications such as object recognition, hand labelling of data is expensive, and there is much interest in semi-supervised techniques based on generative models in which the majority of the training data is unlabelled. Although the generalization performance of generative models can often be improved by 'training them discriminatively', they can then no longer make use of unlabelled data. In an attempt to gain the benefit of both generative and discriminative approaches, heuristic procedure have been proposed [2, 3] which interpolate between these two extremes by taking a convex combination of the generative and discriminative objective functions. In this paper we adopt a new perspective which says that there is only one correct way to train a given model, and that a 'discriminatively trained' generative model is fundamentally a new model [7]. From this viewpoint, generative and discriminative models correspond to specific choices for the prior over parameters. As well as giving a principled interpretation of 'discriminative training', this approach opens door to very general ways of interpolating between generative and discriminative extremes through alternative choices of prior. We illustrate this framework using both synthetic data and a practical example in the domain of multi-class object recognition. Our results show that, when the supply of labelled training data is limited, the optimum performance corresponds to a balance between the purely generative and the purely discriminative.

Dealing with disaster: surviving misbehaved kernel extensions Today's extensible operating systems allow applications to modify kernel behavior by providing mechanisms for application code to run in the kernel address space. The advantage of this approach is that it provides improved application flexibility and performance; the disadvan- tage is that buggy or malicious code can jeopardize the integrity of the kernel. It has been demonstrated that it is feasible to use safe languages, software fault isolation, or virtual memory protection to safeguard the main ker- nel. However, such protection mechanisms do not address the full range of problems, such as resource hoarding, that can arise when application code is intro- duced into the kernel. In this paper, we present an analysis of extension mechanisms in the VINO kernel. VINO uses software fault isolation as its safety mechanism and a lightweight transaction system to cope with resource-hoarding. We explain how these two mechanisms are sufficient to protect against a large class of errant or malicious extensions, and we quantify the overhead that this protection introduces. We find that while the overhead of these techniques is high relative to the cost of the extensions themselves, it is low relative to the benefits that extensibility brings.

Informed mobile prefetching Prefetching is a double-edged sword. It can hide the latency of data transfers over poor and intermittently connected wireless networks, but the costs of prefetching in terms of increased energy and cellular data usage are potentially substantial, particularly for data prefetched incorrectly. Weighing the costs and benefits of prefetching is complex, and consequently most mobile applications employ simple but sub-optimal strategies. Rather than leave the job to applications, we argue that the underlying mobile system should provide explicit prefetching support. Our prototype, IMP, presents a simple interface that hides the complexity of the prefetching decision. IMP uses a cost-benefit analysis to decide when to prefetch data. It employs goal-directed adaptation to try to minimize application response time while meeting budgets for battery lifetime and cellular data usage. IMP opportunistically uses available networks while ensuring that prefetches do not degrade network performance for foreground activity. It tracks hit rates for past prefetches and accounts for network-specific costs in order to dynamically adapt its prefetching strategy to both the network conditions and the accuracy of application prefetch disclosures. Experiments with email and news reader applications show that IMP provides predictable usage of budgeted resources, while lowering application response time compared to the oblivious strategies used by current applications.

Dynamic Knowledge Representation and Its Applications This paper has two main objectives. One is to show that the dynamic knowledge representation paradigm introduced in [ALP+00] and the associated language LUPS, defined in [APPP99], constitute natural, powerful and expressive tools for representing dynamically changing knowledge. We do so by demonstrating the applicability of the dynamic knowledge representation paradigm and the language LUPS to several broad knowledge representation domains, for each of which we provide an illustrative example. Our second objective is to extend our approach to allow proper handling of conflicting updates. So far, our research on knowledge updates was restricted to a two-valued semantics, which, in the presence of conflicting updates, leads to an inconsistent update, even though the updated knowledge base does not necessarily contain any truly contradictory information. By extending our approach to the three-valued semantics we gain the added expressiveness allowing us to express undefined or noncommittal updates.

Rapid Prototyping and Evaluation of Intelligence Functions of Active Storage Devices Active storage devices further improve their performance by executing “intelligence functions,” such as prefetching and data deduplication, in addition to handling the usual I/O requests they receive. Significant research has been carried out to develop effective intelligence functions for the active storage devices. However, laborious and time-consuming efforts are usually required to set up a suitable experimental platform to evaluate each new intelligence function. Moreover, it is difficult to make such prototypes available to other researchers and users to gain valuable experience and feedback. To overcome these difficulties, we propose IOLab, a virtual machine (VM)-based platform for evaluating intelligence functions of active storage devices. The VM-based structure of IOLab enables the evaluation of new (and existing) intelligence functions for different types of OSes and active storage devices with little additional effort. IOLab also supports real-time execution of intelligence functions, providing users opportunities to experience latest intelligence functions without waiting for their deployment in commercial products. Using a set of interesting case studies, we demonstrate the utility of IOLab with negligible performance overhead except for the VM's virtualization overhead.

Impact of multimedia extensions for different processing element granularities on an embedded imaging system Multimedia applications are among the most dominant computing workloads driving innovations in high performance and cost effective systems In this regard, modern general-purpose microprocessors have included multimedia extensions (e.g., MMX, SSE, VIS, MAX, ALTIVEC) to their instruction set architectures to improve the performance of multimedia with little added cost to microprocessors Whereas prior studies of multimedia extensions have primarily focused on a single processor, this paper quantitatively evaluates the impact of multimedia extensions on system performance and efficiency for different number of processing elements (PEs) within an integrated multiprocessor array This paper also identifies the optimal PE granularity for the array system and implementation technology in terms of throughput, area efficiency, and energy efficiency using architectural and workload simulation Experimental results with cycle accurate simulation and technology modeling show that MMX-type instructions (a representative Intel's multimedia extensions) achieve an average speedup ranging from 1.24( (at a 65,536 PE system) to 5.65( (at a 4 PE system) over the baseline performance In addition, the MMX-enhanced processor array increases both area and energy efficiency over the baseline for all the configurations and programs Moreover, the highest area and energy efficiency are achieved at the number of PEs between 256 and 1,024 These evaluation techniques composed of performance simulation and technology modeling can provide solutions to the design challenges in a new class of multiprocessor array systems for multimedia.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Progress in Case-Based Planning Case-based planning (CBP) is an approach to automated planning that tries to save computational effort by reusing previously found solutions. In 2001, Spalazzi published a survey of work in CBP; here, we present an updated overview of systems that have contributed to the evolution of the field or addressed some issues related to planning by reuse in a novel way. The article presents relevant planners so that readers gain insight into the operation of these systems. This analysis will allow readers to understand the approaches both in the quality of the solutions and in the complexity of finding them.

On the complexity of planning for agent teams and its implications for single agent planning If the complexity of planning for a single agent is described by some function f of the input, how much more difficult is it to plan for a team of n cooperating agents? If these agents are completely independent, we can simply solve n single agent problems, scaling linearly with the number of agents. But if all the agents interact tightly, we really need to solve a single problem that is n times larger, which could be exponentially (in n) harder to solve. Is a more general characterization possible? To formulate this question precisely, we minimally extend the standard STRIPS model to describe multi-agent planning problems. Then, we identify two problem parameters that help us answer our question. The first parameter is independent of the precise task the multi-agent system should plan for, and it captures the structure of the possible direct interactions between the agents via the tree-width of a graph induced by the team. The second parameter is task-dependent, and it captures the minimal number of interactions by the ''most interacting'' agent in the team that is needed to solve the problem. We show that multi-agent planning problems can be solved in time exponential only in these parameters. Thus, when these parameters are bounded, the complexity scales only polynomially in the size of the agent team. These results also have direct implications for the single-agent case: by casting single-agent planning tasks as multi-agent planning tasks, we can devise novel methods for decomposition-based planning for single agents. We analyze one such method, and use the techniques developed to provide some of the strongest tractability results for classical single-agent planning to date.

Complexity Results For Sas(+) Planning We have previously reported a number of tractable planning problems defined in the SAS(+) formalism. This article complements these results by providing a complete map over the complexity of SAS(+) planning under all combinations of the previously considered restrictions. We analyze the complexity of both finding a minimal plan and finding any plan. In contrast to other complexity surveys of planning, we study not only the complexity of the decision problems but also the complexity of the generation problems. We prove that the SAS(+)-PUS problem is the maximal tractable problem under the restrictions we have considered if we want to generate minimal plans. If we are satisfied with any plan, then we can generalize further to the SAS(+)-US problem, which we prove to be the maximal tractable problem in this case.

The FF planning system: fast plan generation through heuristic search We describe and evaluate the algorithmic techniques that are used in the FF planning system. Like the HSP system, FF relies on forward state space search, using a heuristic that estimates goal distances by ignoring delete lists. Unlike HSP's heuristic, our method does not assume facts to be independent. We introduce a novel search strategy that combines hill-climbing with systematic search, and we show how other powerful heuristic information can be extracted and used to prune the search space. FF was the most successful automatic planner at the recent AIPS-2000 planning competition. We review the results of the competition, give data for other benchmark domains, and investigate the reasons for the runtime performance of FF compared to HSP.

Logic programs with classical negation

Logic programming and knowledge representation In this paper, we review recent work aimed at the application of declarative logic programming to knowledge representation in artificial intelligence. We consider extensions of the language of definite logic programs by classical (strong) negation, disjunction, and some modal operators and show how each of the added features extends the representational power of the language.

The contract net protocol: high-level communication and control in a distributed problem solver The contract net protocol has been developed to specify problem-solving communication and control for nodes in a distributed problem solver. Task distribution is affected by a negotiation process, a discussion carried on between nodes with tasks to be executed and nodes that may be able to execute those tasks.

A trace-driven analysis of the UNIX 4.2 BSD file system

Synchronized Disk Interleaving A group of disks may be interleaved to speed up data transfers in a manner analogous to the speedup achieved by main memory interleaving. Conventional disks may be used for interleaving by spreading data across disks and by treating multiple disks as if they were a single one. Furthermore, the rotation of the interleaved disks may be synchronized to simplify control and also to optimize performance. In addition, check- sums may be placed on separate check-sum disks in order to improve reliability. In this paper, we study synchronized disk interleaving as a high-performance mass storage system architecture. The advantages and limitations of the proposed disk interleaving scheme are analyzed using the M/G/1 queueing model and compared to the conventional disk access mechanism.

A Completeness Result for SLDNF-Resolution Because of the possibility of floundering and infinite derivations, SLDNF-resolution is, in general, not complete. The classical approach [17] to get a completeness result is to restrict the attention to normal programs P and normal goals G, such that P or {G} is allowed and P is hierarchical. Unfortunately, the class of all normal programs and all normal goals meeting these requirements is not powerful enough to be of great practical importance. But after refining the concept of allowedness by taking modes [12] into account, we can broaden the notion of a hierarchical program, and thereby define a subclass of the class of normal programs and normal goals which is powerful enough to compute all primitive recursive functions without losing the completeness of SLDNF-resolution.

A logic programming approach to knowledge-state planning: Semantics and complexity We propose a new declarative planning language, called K, which is based on principles and methods of logic programming. In this language, transitions between states of knowledge can be described, rather than transitions between completely described states of the world, which makes the language well suited for planning under incomplete knowledge. Furthermore, our formalism enables the use of default principles in the planning process by supporting negation as failure. Nonetheless, K also supports the representation of transitions between states of the world (i.e., states of complete knowledge) as a special case, which shows that the language is very flexible. As we demonstrate on particular examples, the use of knowledge states may allow for a natural and compact problem representation. We then provide a thorough analysis of the computational complexity of K, and consider different planning problems, including standard planning and secure planning (also known as conformant planning) problems. We show that these problems have different complexities under various restrictions, ranging from NP to NEXPTIME in the propositional case. Our results form the theoretical basis for the DLVk system, which implements the language K on top of the DLV logic programming system.

iSAM: Incremental Smoothing and Mapping In this paper, we present incremental smoothing and mapping (iSAM), which is a novel approach to the simultaneous localization and mapping problem that is based on fast incremental matrix factorization. iSAM provides an efficient and exact solution by updating a QR factorization of the naturally sparse smoothing information matrix, thereby recalculating only those matrix entries that actually change. iSAM is efficient even for robot trajectories with many loops as it avoids unnecessary fill-in in the factor matrix by periodic variable reordering. Also, to enable data association in real time, we provide efficient algorithms to access the estimation uncertainties of interest based on the factored information matrix. We systematically evaluate the different components of iSAM as well as the overall algorithm using various simulated and real-world datasets for both landmark and pose-only settings.

When Multivariate Forecasting Meets Unsupervised Feature Learning - Towards a Novel Anomaly Detection Framework for Decision Support. Many organizations adopt information technologies to make intelligent decisions during operations. Time-series data plays a crucial role in supporting such decision making processes. Though current studies on time-series based decision making provide reasonably well results, the anomaly detection essence underling most of the scenarios and the plenitude of unlabeled data are largely overlooked and left unexplored. We argue that by using multivariate forecasting and unsupervised feature learning, these two important research gaps could be filled. We carried out two experiments in this study to testify our approach and the results showed that decision support performance was significantly improved. We also proposed a novel framework to integrate the two methods so that our approach may be generalized to a larger problem domain. We discussed the advantages, the limitations and the future work of our study. Both practical and theoretical contributions were also discussed in the paper. © 2012 by the AIS/ICIS Administrative Office All rights reserved.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Exploring the effect of data reduction on Neural Network and Support Vector Machine classification. Neural Networks and Support Vector Machines (SVMs) are two of the most popular and efficient supervised classification models. However, in the context of large datasets many complexity issues arise due to high memory requirements and high computational cost. In the context of the application of Data Mining algorithms, data reduction techniques attempt to reduce the size of training datasets in terms of the number of instances by selecting some of the existing instances or by generating new training instances. The idea is to speed up the application of the data mining algorithm with minimum or no sacrifice in performance. Data reduction techniques have been extensively used in the context of k-Nearest Neighbor classification, a lazy classifier that works by directly using a training dataset rather than building a model. This paper explores the application of data reduction techniques as a preprocessing step before the training step of Neural Networks and SVMs. Furthermore, the paper proposes a new data reduction technique that is based on k-median clustering algorithm. Our experimental results illustrate that, in the case of SVMs, data reduction techniques can effectively reduce the dataset size incurring small performance degradation. In the case of Neural Networks, the performance loss is somewhat greater, for the same data reduction rate, but both SVM and Neural Network models outperform the k-NN approach that is typically used in Data Mining applications.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Computing Equilibrium Models Using Signed Formulas We discuss equilibrium logic, first presented in Pearce (1997), as a system of nonmonotonic reasoning based on the nonclassical logic N5 of here-and-there with strong negation. Equilibrium logic is a conservative extension of answer set inference, not only for extended, disjunctive logic programs, but also for significant extensions such as the programs with nested expressions described by Lifschitz, Tang and Turner (forthcoming). It provides a theoretical basis for extending the paradigm of answer set programming beyond current systems such as smodels and dlv. The paper provides proof systems for N5 and for model-checking in equilibrium logic. The reduction of the latter problem to an unsatisfiability problem of classical logic yields complexity results for the various decision problems concerning equilibrium entailment. The reduction also yields a basis for the practical implementation of an automated reasoning tool.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

Parameterized complexity for the database theorist

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Coupling weight elimination and genetic algorithms Network size plays an important role in the generalization performance of a network. A number of approaches which try to determine an “appropriate” network size for a given problem have been developed during the last few years. Although it is usually demonstrated that such approaches are capable of finding small size networks that solve the problem at hand, it is quite remarkable that the generalization capabilities of these networks have not been thoroughly explored. In this paper, we have considered the weight elimination technique and we propose a scheme where it is coupled with genetic algorithms. Our objective is not only to find smaller size networks that solve the problem at hand, by pruning larger size networks, but also to improve generalization. The innovation of our work relies on a fitness function which uses an adaptive parameter to encourage the reproduction of networks having good generalization performance and a relatively small size

Improving Generalization of Neural Networks Through Pruning

What Size Net Gives Valid Generalization? We address the question of when a network can be expected to generalize from m random training examples chosen from some arbitrary probability distribution, assuming that future test examples are drawn from the same distribution. Among our results are the following bounds on appropriate sample vs. network size. Assume 0 < ∊ ≤ 1/8. We show that if m ≥ O(W/∊ log N/∊) random examples can be loaded on...

Generalization by weight-elimination with application to forecasting Inspired by the information theoretic idea of minimum description length, we add a term to the back propagation cost function that penalizes network complexity. We give the details of the procedure, called weight-elimination, describe its dynamics, and clarify the meaning of the parameters involved. From a Bayesian perspective, the complexity term can be usefully interpreted as an assumption about prior distribution of the weights. We use this procedure to predict the sunspot time series and the notoriously noisy series of currency exchange rates.

Extended stable semantics for normal and disjunctive programs

A neural probabilistic language model A goal of statistical language modeling is to learn the joint probability function of sequences of words in a language. This is intrinsically difficult because of the curse of dimensionality: a word sequence on which the model will be tested is likely to be different from all the word sequences seen during training. Traditional but very successful approaches based on n-grams obtain generalization by concatenating very short overlapping sequences seen in the training set. We propose to fight the curse of dimensionality by learning a distributed representation for words which allows each training sentence to inform the model about an exponential number of semantically neighboring sentences. The model learns simultaneously (1) a distributed representation for each word along with (2) the probability function for word sequences, expressed in terms of these representations. Generalization is obtained because a sequence of words that has never been seen before gets high probability if it is made of words that are similar (in the sense of having a nearby representation) to words forming an already seen sentence. Training such large models (with millions of parameters) within a reasonable time is itself a significant challenge. We report on experiments using neural networks for the probability function, showing on two text corpora that the proposed approach significantly improves on state-of-the-art n-gram models, and that the proposed approach allows to take advantage of longer contexts.

On the scale and performance of cooperative Web proxy caching Abstract While algorithms for cooperative proxy caching have been widely studied, little is understood about cooperative- caching performance,in the large-scale World Wide Web en- vironment. This paper uses both trace-based analysis and analytic modelling,to show,the potential advantages and drawbacks of inter-proxy cooperation. With our traces, we evaluate quantitatively the performance-improvement po- tential of cooperation between 200 small-organization prox- ies within a university environment, and between two large- organization proxies handling 23,000 and 60,000 clients, re- spectively. With our model, we extend beyond these popula- tions to project cooperative caching behavior in regions with millions of clients. Overall, we demonstrate that cooperative caching has performance,benefits only within limited popu- lation bounds. We also use our model to examine the impli- cations of future trends in Web-access behavior and traffic.

A trace-driven analysis of the UNIX 4.2 BSD file system

Synchronized Disk Interleaving A group of disks may be interleaved to speed up data transfers in a manner analogous to the speedup achieved by main memory interleaving. Conventional disks may be used for interleaving by spreading data across disks and by treating multiple disks as if they were a single one. Furthermore, the rotation of the interleaved disks may be synchronized to simplify control and also to optimize performance. In addition, check- sums may be placed on separate check-sum disks in order to improve reliability. In this paper, we study synchronized disk interleaving as a high-performance mass storage system architecture. The advantages and limitations of the proposed disk interleaving scheme are analyzed using the M/G/1 queueing model and compared to the conventional disk access mechanism.

A Completeness Result for SLDNF-Resolution Because of the possibility of floundering and infinite derivations, SLDNF-resolution is, in general, not complete. The classical approach [17] to get a completeness result is to restrict the attention to normal programs P and normal goals G, such that P or {G} is allowed and P is hierarchical. Unfortunately, the class of all normal programs and all normal goals meeting these requirements is not powerful enough to be of great practical importance. But after refining the concept of allowedness by taking modes [12] into account, we can broaden the notion of a hierarchical program, and thereby define a subclass of the class of normal programs and normal goals which is powerful enough to compute all primitive recursive functions without losing the completeness of SLDNF-resolution.

A logic programming approach to knowledge-state planning: Semantics and complexity We propose a new declarative planning language, called K, which is based on principles and methods of logic programming. In this language, transitions between states of knowledge can be described, rather than transitions between completely described states of the world, which makes the language well suited for planning under incomplete knowledge. Furthermore, our formalism enables the use of default principles in the planning process by supporting negation as failure. Nonetheless, K also supports the representation of transitions between states of the world (i.e., states of complete knowledge) as a special case, which shows that the language is very flexible. As we demonstrate on particular examples, the use of knowledge states may allow for a natural and compact problem representation. We then provide a thorough analysis of the computational complexity of K, and consider different planning problems, including standard planning and secure planning (also known as conformant planning) problems. We show that these problems have different complexities under various restrictions, ranging from NP to NEXPTIME in the propositional case. Our results form the theoretical basis for the DLVk system, which implements the language K on top of the DLV logic programming system.

iSAM: Incremental Smoothing and Mapping In this paper, we present incremental smoothing and mapping (iSAM), which is a novel approach to the simultaneous localization and mapping problem that is based on fast incremental matrix factorization. iSAM provides an efficient and exact solution by updating a QR factorization of the naturally sparse smoothing information matrix, thereby recalculating only those matrix entries that actually change. iSAM is efficient even for robot trajectories with many loops as it avoids unnecessary fill-in in the factor matrix by periodic variable reordering. Also, to enable data association in real time, we provide efficient algorithms to access the estimation uncertainties of interest based on the factored information matrix. We systematically evaluate the different components of iSAM as well as the overall algorithm using various simulated and real-world datasets for both landmark and pose-only settings.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

NCSU_SAS_SAM: Deep Encoding and Reconstruction for Normalization of Noisy Text As a participant in the W-NUT Lexical Normalization for English Tweets challenge, we use deep learning to address the constrained task. Specifically, we use a combination of two augmented feed forward neural networks, a flagger that identifies words to be normalized and a normalizer, to take in a single token at a time and output a corrected version of that token. Despite avoiding off-the-shelf tools trained on external data and being an entirely context-free model, our system still achieved an F1-score of 81.49%, comfortably surpassing the next runner up by 1.5% and trailing the second place model by only 0.26%.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Augmenting Supervised Neural Networks with Unsupervised Objectives for Large-scale Image Classification. Unsupervised learning and supervised learning are key research topics in deep learning. However, as high-capacity supervised neural networks trained with a large amount of labels have achieved remarkable success in many computer vision tasks, the availability of large-scale labeled images reduced the significance of unsupervised learning. Inspired by the recent trend toward revisiting the importance of unsupervised learning, we investigate joint supervised and unsupervised learning in a large-scale setting by augmenting existing neural networks with decoding pathways for reconstruction. First, we demonstrate that the intermediate activations of pretrained large-scale classification networks preserve almost all the information of input images except a portion of local spatial details. Then, by end-to-end training of the entire augmented architecture with the reconstructive objective, we show improvement of the network performance for supervised tasks. We evaluate several variants of autoencoders, including the recently proposed what-where autoencoder that uses the encoder pooling switches, to study the importance of the architecture design. Taking the 16-layer VGGNet trained under the ImageNet ILSVRC 2012 protocol as a strong baseline for image classification, our methods improve the validation-set accuracy by a noticeable margin.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Proactive Serving Decreases User Delay Exponentially In online service systems, the delay experienced by a user from the service request to the service completion is one of the most critical performance metrics. To improve user delay experience, recent industrial practice suggests a modern system design mechanism: proactive serving, where the system predicts future user requests and allocates its capacity to serve these upcoming requests proactively. In this paper, we investigate the fundamentals of proactive serving from a theoretical perspective. In particular, we show that proactive serving decreases average delay exponentially (as a function of the prediction window size). Our results provide theoretical foundations for proactive serving and shed light on its application in practical systems.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

A Semantic Web Services Approach Towards Automated Software Engineering The growing complexity of software, combined with demands for greater productivity and shorter cycles, creates an increasing demand for more automation and integration within the software engineering (SE) domain. When viewed holistically, the heterogeneous nature, implicit feature cross-dependencies, and manual administration of the toolchain infrastructure results in unnecessary complexity, inefficiencies, and reduced reliability for the SE process. A common infrastructure is missing that provides an interoperable and distributed tool environment, addresses feature dependency selection, and automates toolchain workflow composition and execution. To address these challenges, this paper explores the practicality of a unifying Semantic Web Services approach towards Automated Software Engineering (SWS-ASE).

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Logic and Databases: Past, Present, and Future

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Performance based design of high-level language-directed computerarchitectures This paper is concerned with the analytical modeling of computer architectures to aid in the design of high-level language-directed computer architectures. High-level language-directed computers are computers that execute programs in a high-level language directly. The design procedure of these computers are at best described as being ad hoc. In order to systematize the design procedure, we introduce analytical models of computers that predict the performance of parallel computations on concurrent computers. We model computers as queueing networks and parallel computations as precedence graphs. The models that we propose are simple and lead to computationally efficient procedures of predicting the performance of parallel computations on concurrent computers. We demonstrate the use of these models in the design of high-level language-directed computer architectures

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

SPIFFI-A Scalable Parallel File System for the Intel Paragon This paper presents the design and performance of SPIFFI, a scalable high-performance parallel file system intended for use by extremely I/O intensive applications including "Grand Challenge" scientific applications and multimedia systems. This paper contains experimental results from a SPIFFI prototype on a 64 node/64 disk Intel Paragon. The results show that SPIFFI provides high performance and linear scaleup on real hardware. The paper also explains how shared file pointers (i.e., file pointers that are shared by multiple processes) can simplify the design of a parallel application. By sequentializing I/O accesses and by providing dynamic I/O load balancing, a shared file pointer may even improve an application's performance.This paper also presents the predictions of a SPIFFI simulator that we validated using the prototype. The simulator results show that SPIFFI continues to provide high performance even when it is scaled to configurations with as many as 128 disks or 256 compute nodes.

Heuristics for Scheduling I/O Operations The I/O bottleneck in parallel computer systems has recently begun receiving increasing interest. Most attention has focused on improving the performance of I/O devices using fairly low-level parallelism in techniques such as disk striping and interleaving. Widely applicable solutions, however, will require an integrated approach which addresses the problem at multiple system levels, including applications, systems software, and architecture. We propose that within the context of such an integrated approach, scheduling parallel I/O operations will become increasingly attractive and can potentially provide substantial performance benefits.We describe a simple I/O scheduling problem and present approximate algorithms for its solution. The costs of using these algorithms in terms of execution time, and the benefits in terms of reduced time to complete a batch of I/O operations, are compared with the situations in which no scheduling is used, and in which an optimal scheduling algorithm is used. The comparison is performed both theoretically and experimentally. We have found that, in exchange for a small execution time overhead, the approximate scheduling algorithms can provide substantial improvements in I/O completion times.

HFS: a performance-oriented flexible file system based on building-block compositions The Hurricane File System (HFS) is designed for (potentially large-scale) shared-memory multiprocessors. Its architecture is based on the principle that, in order to maximize performance for applications with diverse requirements, a file system must support a wide variety of file structures, file system policies, and I/O interfaces. Files in HFS are implemented using simple building blocks composed in potentially complex ways. This approach yields great flexibility, allowing an application to customize the structure and policies of a file to exactly meet its requirements. As an extreme example, HFS allows a file's structure to be optimized for concurrent random-access write-only operations by 10 threads, something no other file system can do. Similarly, the prefetching, locking, and file cache management policies can all be chosen to match an application's access pattern. In contrast, most parallel file systems support a single file structure and a small set of policies. We have implemented HFS as part of the Hurricane operating system running on the Hector shared-memory multiprocessor. We demonstrate that the flexibility of HFS comes with little processing or I/O overhead. We also show that for a number of file access patterns, HFS is able to deliver to the applications the full I/O bandwidth of the disks on our system.

SWIFT: USING DISTRIBUTED DISK STRIPING TO PROVIDE HIGH I/O DATA RATES We present an I/O architecture, called Swift, that addresses the problem of data rate mismatches between the requirements of an application, storage devices, and the interconnection medium. The goal of Swift is to support high data rates in general purpose distributed systems. Swift uses a high-speed interconnection medium to provide high data rate transfers by using multiple slower storage devices in parallel. It scales well when using multiple storage devices and interconnections, and can use any appropriate storage technology, including high-performance devices such as disk arrays. To address the problem of partial failures, Swift stores data redundantly. Using the UNIX operating system, we have constructed a simplified prototype of the Swift architecture. The prototype provides data rates that are significantly faster than access to the local SCSI disk, limited by the capacity of a single Ethernet segment, or in the case of multiple Ethernet segments by the ability of the client to drive them. We have constructed a simulation model to demonstrate how the Swift architecture can exploit advances in processor, communication and storage technology. We consider the effects of processor speed, interconnection capacity, and multiple storage agents on the utilization of the components and the data rate of the system. We show that the data rates scale well in the number of storage devices, and that by replacing the most highly stressed components by more powerful ones the data rates of the entire system increase significantly.

Staggered Striping in Multimedia Information Systems Multimedia information systems have emerged as an essential component of many application domains ranging from library information systems to entertainment technology. However, most implementations of these systems cannot support the continuous display of multimedia objects and suffer from frequent disruptions and delays termed hiccups. This is due to the low I/O bandwidth of the current disk technology, the high bandwidth requirement of multimedia objects, and the large size of these objects that almost always requires them to be disk resident. One approach to resolve this limitation is to decluster a multimedia object across multiple disk drives in order to employ the aggregate bandwidth of several disks to support the continuous retrieval (and display) of objects. This paper describes staggered striping as a novel technique to provide effective support for multiple users accessing the different objects in the database. Detailed simulations confirm the superiority of staggered striping.

The Tiger Shark file system Tiger Shark is a parallel file system for IBM's AIX operating system. It is designed to support interactive multimedia, particularly large-scale systems such as interactive television (ITV). Tiger Shark scales across the entire RS/6000 product line, from small desktop machines to the SP-2 parallel supercomputer. Tiger Shark's primary features are support for continuous time data, scalability, high availability, and manageability, all of which are crucial in its role in large-scale video servers. Interestingly, most of the features that make Tiger Shark a good video server are important for other large-scale applications such as technical computing, data mining, digital library, and scalable network file servers. This paper briefly describes Tiger Shark: the environment that makes it important, the key technology it embodies, and the efforts to build products based on it.

Virtual log based file systems for a programmable disk In this paper, we study how to minimize the latency of small transactional writes to disk. The basic approach is to write to free sectors that are near the current disk head location by leveraging the embedded processor core inside the disk. We develop a number of analytical models to demonstrate the performance potential of this approach. We then present the design of a variation of a log-structured file system based on the concept of a virtual log, which supports fast small transactional writes without extra hardware support. We compare our approach against traditional update-in-place and logging systems by modifying the Solaris kernel to serve as a simulation engine. Our evaluations show that random synchronous updates on an unmodified UFS execute up to an order of magnitude faster on a virtual log than on a conventional disk. The virtual log can also significantly improve LFS in cases where delaying small writes is not an option or on-line cleaning would degrade performance. If the current trends of disk technology continue, we expect the performance advantage of this approach to become even more pronounced in the future.

Maximizing performance in a striped disk array Improvements in disk speeds have not kept up with improvements in processor and memory speeds. One way to correct the resulting speed mismatch is to stripe data across many disks. The authors address how to stripe data to get maximum performance from the disks. Specifically, they examine how to choose the striping unit, that is, the amount of logically contiguous data on each disk. Rules for determining the best striping unit for a given range of workloads are synthesized. It is shown how the choice of striping unit depends on only two parameters: (1) the number of outstanding requests in the disk system at any given time, and (2) the average positioning time×data transfer rate of the disks. The authors derive an equation for the optimal striping unit as a function of these two parameters; they also show how to choose the striping unit without prior knowledge about the workload

The DASDBS Project: Objectives, Experiences, and Future Prospects A retrospective of the Darmstadt database system project, also known as DASDBS, is presented. The project is aimed at providing data management support for advanced applications, such as geo-scientific information systems and office automation. Similar to the dichotomy of RSS and RDS in System R, a layered architectural approach was pursued: a storage management kernel serves as the lowest common denominator of the requirements of the various applications classes, and a family of application-oriented front-ends provides semantically richer functions on top of the kernel. The lessons that were learned from building the DASDBS system are discussed. Particular emphasis is placed on the following issues: the role of nested relations, the experiences with using object buffers for coupling the system with the programming-language environment and the learning process in implementing multilevel transactions.

A distributed file service based on optimistic concurrency control The design of a layered file service for the Amoeba Distributed System is discussed, on top of which various applications can easily be intplemented. The bottom layer is formed by the Amoeba Block Services, responsible for implementing stable storage and repficated, highly available disk blocks. The next layer is formed by the Amoeba File Service which provides version management and concur~ncy control for tree-structured files. On top of this layer, the appficafions, ranging from databases to source code control systems, determine the structure of the file trees and provide an interface to the users.

On the facial structure of set packing polyhedra In this paper we address ourselves to identifying facets of the set packing polyhedron, i.e., of the convex hull of integer solutions to the set covering problem with equality constraints and/or constraints of the form “?”. This is done by using the equivalent node-packing problem derived from the intersection graph associated with the problem under consideration. First, we show that the cliques of the intersection graph provide a first set of facets for the polyhedron in question. Second, it is shown that the cycles without chords of odd length of the intersection graph give rise to a further set of facets. A rather strong geometric property of this set of facets is exhibited.

Random search for hyper-parameter optimization Grid search and manual search are the most widely used strategies for hyper-parameter optimization. This paper shows empirically and theoretically that randomly chosen trials are more efficient for hyper-parameter optimization than trials on a grid. Empirical evidence comes from a comparison with a large previous study that used grid search and manual search to configure neural networks and deep belief networks. Compared with neural networks configured by a pure grid search, we find that random search over the same domain is able to find models that are as good or better within a small fraction of the computation time. Granting random search the same computational budget, random search finds better models by effectively searching a larger, less promising configuration space. Compared with deep belief networks configured by a thoughtful combination of manual search and grid search, purely random search over the same 32-dimensional configuration space found statistically equal performance on four of seven data sets, and superior performance on one of seven. A Gaussian process analysis of the function from hyper-parameters to validation set performance reveals that for most data sets only a few of the hyper-parameters really matter, but that different hyper-parameters are important on different data sets. This phenomenon makes grid search a poor choice for configuring algorithms for new data sets. Our analysis casts some light on why recent "High Throughput" methods achieve surprising success--they appear to search through a large number of hyper-parameters because most hyper-parameters do not matter much. We anticipate that growing interest in large hierarchical models will place an increasing burden on techniques for hyper-parameter optimization; this work shows that random search is a natural baseline against which to judge progress in the development of adaptive (sequential) hyper-parameter optimization algorithms.

S/390 CMOS server I/O: The continuing evolution IBM has developed a strategy to achieve the high I/O demands of large servers. In a new environment of industry-standard peripheral component interconnect (PCI) attached adapters conforming to open I/O interfaces, S/390® has developed an efficient method of quickly integrating disk storage, communications, and future adapters. Preserving the S/390 I/O programming model and the high level of data integrity expected in S/390 products and reducing development cycle time and resources have further constrained design options. At the same time, S/390 developers have redesigned the traditional I/O components into the latest chip technologies. The developers have also designed a new internal link (STI) to meet the increased I/O bandwidth and connectivity required by the high processor performance of the third and fourth generations of S/390 CMOS servers. This paper describes this strategy and how it has led to systems that retain the differentiating features of S/390 products.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

SNPHarvester: a filtering-based approach for detecting epistatic interactions in genome-wide association studies. Motivation: Hundreds of thousands of single nucleotide polymorphisms (SNPs) are available for genome-wide association (GWA) studies nowadays. The epistatic interactions of SNPs are believed to be very important in determining individual susceptibility to complex diseases. However, existing methods for SNP interaction discovery either suffer from high computation complexity or perform poorly when marginal effects of disease loci are weak or absent. Hence, it is desirable to develop an effective method to search epistatic interactions in genome-wide scale. Results: We propose a new method SNPHarvester to detect SNP SNP interactions in GWA studies. SNPHarvester creates multiple paths in which the visited SNP groups tend to be statistically associated with diseases, and then harvests those significant SNP groups which pass the statistical tests. It greatly reduces the number of SNPs. Consequently, existing tools can be directly used to detect epistatic interactions. By using a wide range of simulated data and a real genome-wide data, we demonstrate that SNPHarvester outperforms its recent competitor significantly and is promising for practical disease prognosis.

TEAM: efficient two-locus epistasis tests in human genome-wide association study. As a promising tool for identifying genetic markers underlying phenotypic differences, genome-wide association study (GWAS) has been extensively investigated in recent years. In GWAS, detecting epistasis (or gene-gene interaction) is preferable over single locus study since many diseases are known to be complex traits. A brute force search is infeasible for epistasis detection in the genomewide scale because of the intensive computational burden. Existing epistasis detection algorithms are designed for dataset consisting of homozygous markers and small sample size. In human study, however, the genotype may be heterozygous, and number of individuals can be up to thousands. Thus, existing methods are not readily applicable to human datasets. In this article, we propose an efficient algorithm, TEAM, which significantly speeds up epistasis detection for human GWAS. Our algorithm is exhaustive, i.e. it does not ignore any epistatic interaction. Utilizing the minimum spanning tree structure, the algorithm incrementally updates the contingency tables for epistatic tests without scanning all individuals. Our algorithm has broader applicability and is more efficient than existing methods for large sample study. It supports any statistical test that is based on contingency tables, and enables both family-wise error rate and false discovery rate controlling. Extensive experiments show that our algorithm only needs to examine a small portion of the individuals to update the contingency tables, and it achieves at least an order of magnitude speed up over the brute force approach.

Predictive rule inference for epistatic interaction detection in genome-wide association studies. Under the current era of genome-wide association study (GWAS), finding epistatic interactions in the large volume of SNP data is a challenging and unsolved issue. Few of previous studies could handle genome-wide data due to the difficulties in searching the combinatorially explosive search space and statistically evaluating high-order epistatic interactions given the limited number of samples. In this work, we propose a novel learning approach (SNPRuler) based on the predictive rule inference to find disease-associated epistatic interactions.Our extensive experiments on both simulated data and real genome-wide data from Wellcome Trust Case Control Consortium (WTCCC) show that SNPRuler significantly outperforms its recent competitor. To our knowledge, SNPRuler is the first method that guarantees to find the epistatic interactions without exhaustive search. Our results indicate that finding epistatic interactions in GWAS is computationally attainable in practice.http://bioinformatics.ust.hk/SNPRuler.zip

Travelling the world of gene-gene interactions. Over the last few years, main effect genetic association analysis has proven to be a successful tool to unravel genetic risk components to a variety of complex diseases. In the quest for disease susceptibility factors and the search for the 'missing heritability', supplementary and complementary efforts have been undertaken. These include the inclusion of several genetic inheritance assumptions in model development, the consideration of different sources of information, and the acknowledgement of disease underlying pathways of networks. The search for epistasis or gene-gene interaction effects on traits of interest is marked by an exponential growth, not only in terms of methodological development, but also in terms of practical applications, translation of statistical epistasis to biological epistasis and integration of omics information sources. The current popularity of the field, as well as its attraction to interdisciplinary teams, each making valuable contributions with sometimes rather unique viewpoints, renders it impossible to give an exhaustive review of to-date available approaches for epistasis screening. The purpose of this work is to give a perspective view on a selection of currently active analysis strategies and concerns in the context of epistasis detection, and to provide an eye to the future of gene-gene interaction analysis.

Improvement of BLASTp on the FPGA-Based high-performance computer RIVYERA NCBI BLASTp plays the major role of protein database searches already for years. However, with today's growth of sequence database sizes, it becomes more inefficient with standard PC architectures. One solution to address this problem was already presented in our previous implementation, published in [16], taking advantages of the massive parallelization provided by the FPGA-based high-performance computer RIVYERA [3]. The analysis of bottlenecks in our BLASTp pipeline showed the urgent need to speed up the two-hit finder component, as well as the postprocessing on the PC. After a complete redesign of the two-hit finder and the insertion of a new "gapped extension" filter, we achieve a speedup of up to 376, compared to one thread of a fully utilized 2x Intel Xeon E5520 PC system at $2.26\ensuremath{\mathrm{GHz}} $ running original NCBI BLASTp v. 2.2.25+. This is about two times the performance of our previous implementation.

Design of a FPGA-Based Parallel Architecture for BLAST Algorithm with Multi-hits Detection In this paper, a design of a Field Programmable Gate Array (FPGA)-based parallel architecture for Basic Local Alignment Search Tool (BLAST) Algorithm with multi-hits Detection is proposed. Basic Local Alignment Search Tool algorithm is a heuristic biological sequence alignment algorithm and has been widely used in the computational biology domain. The architecture contains several different blocks. Each block will perform a different step of National Center for Biotechnology Information (NCBI) BLAST family algorithms in parallel. The most appealing and distinguishing features of this architecture are the Multiple Hits Finder Array and Hits Combination Block. In the hits detection stage, we apply a Multiple Hits Finder Array approach to realize the multi-hits in one clock cycle. Then the multiple overlapping hits are merged together in the Hits Finder Block to reduce memory consumption and save implementation time. The parallel design makes this FPGA-based BLAST algorithm implementation much faster than an equivalent software program would in a PC. Multiple Hits Finder Array architecture prototype is implemented and the storage expense and synthesis performance of the prototype are discussed.

Bioinformatics Research and Applications: 8th International Symposium, ISBRA 2012, Dallas, TX, USA, May 21-23, 2012. Proceedings

Accelerating BLASTP on the Cell Broadband Engine The enormous growth of biological sequence databases has caused bioinformatics to be rapidly moving towards a data-intensive, computational science. As a result, the computational power needed by bioinformatics applications is growing rapidly as well. The recent emergence of low cost parallel accelerator technologies has made it possible to reduce execution times of many bioinformatics applications. In this paper, we demonstrate how the PlayStation®3, powered by the Cell Broadband Engine, can be used as an efficient computational platform to accelerate the popular BLASTP algorithm.

The HP AutoRAID hierarchical storage system Configuring redundant disk arrays is a black art. To configure an array properly, a system administrator must understand the details of both the array and the workload it will support. Incorrect understanding of either, or changes in the workload over time, can lead to poor performance. We present a solution to this problem: a two-level storage hierarchy implemented inside a single disk-array controller. In the upper level of this hierarchy, two copies of active data are stored to provide full redundancy and excellent performance. In the lower level, RAID 5 parity protection is used to provide excellent storage cost for inactive data, at somewhat lower performance. The technology we describe in this article, know as HP AutoRAID, automatically and transparently manages migration of data blocks between these two levels as access patterns change. The result is a fully redundant storage system that is extremely easy to use, is suitable for a wide variety of workloads, is largely insensitive to dynamic workload changes, and performs much better than disk arrays with comparable numbers of spindles and much larger amounts of front-end RAM cache. Because the implementation of the HP AutoRAID technology is almost entirely in software, the additional hardware cost for these benefits is very small. We describe the HP AutoRAID technology in detail, provide performance data for an embodiment of it in a storage array, and summarize the results of simulation studies used to choose algorithms implemented in the array.

Perceiving And Reasoning About A Changing World A rational agent (artificial or otherwise) residing in a complex changing environment must gather information perceptually, update that information as the world changes, and combing that information with causal information to reason about the changing world. Using the system of defeasible reasoning that is incorporated into the OSCAR architecture for rational agents, a set of reason-schemas is proposed for enabling an agent to perform some of the requisite reasoning. Along the way, solutions are proposed for the Frame Problem, the Qualification Problem, and the Ramification Problem. The principles and reasoning described have all been implemented in OSCAR.

An overview of MetaMap: historical perspective and recent advances. MetaMap is a widely available program providing access to the concepts in the unified medical language system (UMLS) Metathesaurus from biomedical text. This study reports on MetaMap's evolution over more than a decade, concentrating on those features arising out of the research needs of the biomedical informatics community both within and outside of the National Library of Medicine. Such features include the detection of author-defined acronyms/abbreviations, the ability to browse the Metathesaurus for concepts even tenuously related to input text, the detection of negation in situations in which the polarity of predications is important, word sense disambiguation (WSD), and various technical and algorithmic features. Near-term plans for MetaMap development include the incorporation of chemical name recognition and enhanced WSD.

Variable minimal unsatisfiability In this paper, we present variable minimal unsatisfiability (VMU), which is a generalization of minimal unsatisfiability (MU). A characterization of a VMU formula F is that every variable of F is used in every resolution refutation of F. We show that the class of VMU formulas is DP-complete. For fixed deficiency (the difference of the number of clauses and the number of variables), the VMU formulas can be solved in polynomial time. Furthermore, we investigate more subclasses of VMU formulas. Although the theoretic results on VMU and MU are similar, some observations are shown that the extraction of VMU may be more practical than MU in some cases.

Evolving mach 3.0 to a migrating thread model We have modified Mach 3.0 to treat cross-domain remote procedure call (RPC) as a single entity, instead of a sequence of message passing operations. With RPC thus elevated, we improved the transfer of control during RPC by changing the thread model. Like most operating systems, Mach views threads as statically associated with a single task, with two threads involved in an RPC. An alternate model is that of migrating threads, in which, during RPC, a single thread abstraction moves between tasks with the logical flow of control, and "server" code is passively executed. We have compatibly replaced Mach's static threads with migrating threads, in an attempt to isolate this aspect of operating system design and implementation. The key element of our design is a decoupling of the thread abstraction into the execution context and the schedulable thread of control, consisting of a chain of contexts. A key element of our implementation is that threads are now "based" in the kernel, and temporarily make excursions into tasks via upcalls. The new system provides more precisely defined semantics for thread manipulation and additional control operations, allows scheduling and accounting attributes to follow threads, simplifies kernel code, and improves RPC performance. We have retained the old thread and IPC interfaces for backwards compatibility, with no changes required to existing client programs and only a minimal change to servers, as demonstrated by a functional Unix single server and clients. The logical complexity along the critical RPC path has been reduced by a factor of nine. Local RPC, doing normal marshaling, has sped up by factors of 1.7-3.4. We conclude that a migrating-thread model is superior to a static model, that kernel-visible RPC is a prerequisite for this improvement, and that it is feasible to improve existing operating systems in this manner.

Learning A Lexical Simplifier Using Wikipedia In this paper we introduce a new lexical simplification approach. We extract over 30K candidate lexical simplifications by identifying aligned words in a sentence-aligned corpus of English Wikipedia with Simple English Wikipedia. To apply these rules, we learn a feature-based ranker using SVMnk trained on a set of labeled simplifications collected using Amazon's Mechanical Turk. Using human simplifications for evaluation, we achieve a precision of 76% with changes in 86% of the examples.

Learned Binary Spectral Shape Descriptor For 3d Shape Correspondence Dense 3D shape correspondence is an important problem in computer vision and computer graphics. Recently, the local shape descriptor based 3D shape correspondence approaches have been widely studied, where the local shape descriptor is a real-valued vector to characterize the geometrical structure of the shape. Different from these real-valued local shape descriptors, in this paper, we propose to learn a novel binary spectral shape descriptor with the deep neural network for 3D shape correspondence. The binary spectral shape descriptor can require less storage space and enable fast matching. First, based on the eigen-vectors of the Laplace-Beltrami operator, we construct a neural network to form a nonlinear spectral representation to characterize the shape. Then, for the defined positive and negative points on the shapes, we train the constructed neural network by minimizing the errors between the outputs and their corresponding binary descriptors, minimizing the variations of the outputs of the positive points and maximizing the variations of the outputs of the negative points, simultaneously. Finally, we binarize the output of the neural network to form the binary spectral shape descriptor for shape correspondence. The proposed binary spectral shape descriptor is evaluated on the SCAPE and TOSCA 3D shape datasets for shape correspondence. The experimental results demonstrate the effectiveness of the proposed binary shape descriptor for the shape correspondence task.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Big Neural Networks Waste Capacity This article exposes the failure of some big neural networks to leverage added capacity to reduce underfitting. Past research suggest diminishing returns when increasing the size of neural networks. Our experiments on ImageNet LSVRC-2010 show that this may be due to the fact there are highly diminishing returns for capacity in terms of training error, leading to underfitting. This suggests that the optimization method - first order gradient descent - fails at this regime. Directly attacking this problem, either through the optimization method or the choices of parametrization, may allow to improve the generalization error on large datasets, for which a large capacity is required.

Evolving Culture vs Local Minima We propose a theory that relates difficulty of learning in deep architectures to culture and language. It is articulated around the following hypotheses: (1) learning in an individual human brain is hampered by the presence of effective local minima; (2) this optimization difficulty is particularly important when it comes to learning higher-level abstractions, i.e., concepts that cover a vast and highly-nonlinear span of sensory configurations; (3) such high-level abstractions are best represented in brains by the composition of many levels of representation, i.e., by deep architectures; (4) a human brain can learn such high-level abstractions if guided by the signals produced by other humans, which act as hints or indirect supervision for these high-level abstractions; and (5), language and the recombination and optimization of mental concepts provide an efficient evolutionary recombination operator, and this gives rise to rapid search in the space of communicable ideas that help humans build up better high-level internal representations of their world. These hypotheses put together imply that human culture and the evolution of ideas have been crucial to counter an optimization difficulty: this optimization difficulty would otherwise make it very difficult for human brains to capture high-level knowledge of the world. The theory is grounded in experimental observations of the difficulties of training deep artificial neural networks. Plausible consequences of this theory for the efficiency of cultural evolutions are sketched.

Estimating or Propagating Gradients Through Stochastic Neurons Stochastic neurons can be useful for a number of reasons in deep learning models, but in many cases they pose a challenging problem: how to estimate the gradient of a loss function with respect to the input of such stochastic neurons, i.e., can we "back-propagate" through these stochastic neurons? We examine this question, existing approaches, and present two novel families of solutions, applicable in different settings. In particular, it is demonstrated that a simple biologically plausible formula gives rise to an an unbiased (but noisy) estimator of the gradient with respect to a binary stochastic neuron firing probability. Unlike other estimators which view the noise as a small perturbation in order to estimate gradients by finite differences, this estimator is unbiased even without assuming that the stochastic perturbation is small. This estimator is also interesting because it can be applied in very general settings which do not allow gradient back-propagation, including the estimation of the gradient with respect to future rewards, as required in reinforcement learning setups. We also propose an approach to approximating this unbiased but high-variance estimator by learning to predict it using a biased estimator. The second approach we propose assumes that an estimator of the gradient can be back-propagated and it provides an unbiased estimator of the gradient, but can only work with non-linearities unlike the hard threshold, but like the rectifier, that are not flat for all of their range. This is similar to traditional sigmoidal units but has the advantage that for many inputs, a hard decision (e.g., a 0 output) can be produced, which would be convenient for conditional computation and achieving sparse representations and sparse gradients.

Knowledge Matters: Importance of Prior Information for Optimization We explored the effect of introducing prior knowledge into the intermediate level of deep supervised neural networks on two tasks. On a task we designed, all black-box state-of-theart machine learning algorithms which we tested, failed to generalize well. We motivate our work from the hypothesis that, there is a training barrier involved in the nature of such tasks, and that humans learn useful intermediate concepts from other individuals by using a form of supervision or guidance using a curriculum. Our results provide a positive evidence in favor of this hypothesis. In our experiments, we trained a two-tiered MLP architecture on a dataset for which each input image contains three sprites, and the binary target class is 1 if all of three shapes belong to the same category and otherwise the class is 0. In terms of generalization, black-box machine learning algorithms could not perform better than chance on this task. Standard deep supervised neural networks also failed to generalize. However, using a particular structure and guiding the learner by providing intermediate targets in the form of intermediate concepts (the presence of each object) allowed us to solve the task efficiently. We obtained much better than chance, but imperfect results by exploring different architectures and optimization variants. This observation might be an indication of optimization difficulty when the neural network trained without hints on this task. We hypothesize that the learning difficulty is due to the composition of two highly non-linear tasks. Our findings are also consistent with the hypotheses on cultural learning inspired by the observations of training of neural networks sometimes getting stuck, even though good solutions exist, both in terms of training and generalization error.

Differentiable Sparse Coding Prior work has shown that features which appear to be biologically plausible as well as empirically useful can be found by sparse coding with a prior such as a laplacian (L1) that promotes sparsity. We show how smoother priors can pre- serve the benefits of these sparse priors while adding stability to the Maximum A-Posteriori (MAP) estimate that makes it more useful for prediction problems. Additionally, we show how to calculate the derivative of the MAP estimate effi- ciently with implicit differentiation. One prior that can be differentiated this way is KL-regularization. We demonstrate its effectiveness on a wide variety of appli- cations, and find that online optimization of the parameters of the KL-regularized model can significantly improve prediction performance.

Better Mixing via Deep Representations It has previously been hypothesized, and supported with some experimental evidence, that deeper representations, when well trained, tend to do a better job at disentangling the underlying factors of variation. We study the following related conjecture: better representations, in the sense of better disentangling, can be exploited to produce faster-mixing Markov chains. Consequently, mixing would be more efficient at higher levels of representation. To better understand why and how this is happening, we propose a secondary conjecture: the higher-level samples fill more uniformly the space they occupy and the high-density manifolds tend to unfold when represented at higher levels. The paper discusses these hypotheses and tests them experimentally through visualization and measurements of mixing and interpolating between samples.

Learning long-term dependencies with gradient descent is difficult Recurrent neural networks can be used to map input sequences to output sequences, such as for recognition, production or prediction problems. However, practical difficulties have been reported in training recurrent neural networks to perform tasks in which the temporal contingencies present in the input/output sequences span long intervals. We show why gradient based learning algorithms face an increasingly difficult problem as the duration of the dependencies to be captured increases. These results expose a trade-off between efficient learning by gradient descent and latching on information for long periods. Based on an understanding of this problem, alternatives to standard gradient descent are considered.

Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning. Deep Learning has recently become hugely popular in machine learning for its ability to solve end-to-end learning systems, in which the features and the classifiers are learned simultaneously, providing significant improvements in classification accuracy in the presence of highly-structured and large databases. Its success is due to a combination of recent algorithmic breakthroughs, increasingly powerful computers, and access to significant amounts of data. Researchers have also considered privacy implications of deep learning. Models are typically trained in a centralized manner with all the data being processed by the same training algorithm. If the data is a collection of users' private data, including habits, personal pictures, geographical positions, interests, and more, the centralized server will have access to sensitive information that could potentially be mishandled. To tackle this problem, collaborative deep learning models have recently been proposed where parties locally train their deep learning structures and only share a subset of the parameters in the attempt to keep their respective training sets private. Parameters can also be obfuscated via differential privacy (DP) to make information extraction even more challenging, as proposed by Shokri and Shmatikov at CCS'15. Unfortunately, we show that any privacy-preserving collaborative deep learning is susceptible to a powerful attack that we devise in this paper. In particular, we show that a distributed, federated, or decentralized deep learning approach is fundamentally broken and does not protect the training sets of honest participants. The attack we developed exploits the real-time nature of the learning process that allows the adversary to train a Generative Adversarial Network (GAN) that generates prototypical samples of the targeted training set that was meant to be private (the samples generated by the GAN are intended to come from the same distribution as the training data). Interestingly, we show that record-level differential privacy applied to the shared parameters of the model, as suggested in previous work, is ineffective (i.e., record-level DP is not designed to address our attack).

On the quantitative analysis of deep belief networks Deep Belief Networks (DBN's) are generative models that contain many layers of hidden variables. Efficient greedy algorithms for learning and approximate inference have allowed these models to be applied successfully in many application domains. The main building block of a DBN is a bipartite undirected graphical model called a restricted Boltzmann machine (RBM). Due to the presence of the partition function, model selection, complexity control, and exact maximum likelihood learning in RBM's are intractable. We show that Annealed Importance Sampling (AIS) can be used to efficiently estimate the partition function of an RBM, and we present a novel AIS scheme for comparing RBM's with different architectures. We further show how an AIS estimator, along with approximate inference, can be used to estimate a lower bound on the log-probability that a DBN model with multiple hidden layers assigns to the test data. This is, to our knowledge, the first step towards obtaining quantitative results that would allow us to directly assess the performance of Deep Belief Networks as generative models of data.

A Provably Efficient Algorithm for Training Deep Networks

Performance Analysis of RAID5 Disk Arrays with a Vacationing Server Model for Rebuild Mode Operation We analyze the performance of RAIDS disk arrays in normal, degraded, and rebuild modes. The analysis, which is shown to be highly accurate through validation against simulation results, achieves its accuracy by (1) modeling detailed disk characteristics; (2) developing a simple approximation to compute the mean response time for fork-join requests arising in degraded mode operation; and (3) using a vacationing server model with multiple vacation types for rebuild mode analysis. According to this model vacations (rebuild reads) are started when the server (disk) becomes idle and are repeated until the arrival of an external disk request. Type one (two) vacations correspond to the reading of the first track which requires a seek (successive tracks requiring no seeks). The analytic solution is used to quantify the effect of different rebuild options, such as read redirection and the split-seek option

Selected topics on assignment problems We survey recent developments in the fields of bipartite matchings, linear sum assignment and bottleneck assignment problems and applications, multidimensional assignment problems, quadratic assignment problems, in particular lower bounds, special cases and asymptotic results, biquadratic and communication assignment problems.

When Are Behaviour Networks Well-Behaved? Agents operating in the real world have to deal with a constantly changing and only partially predictable environment and are nevertheless expected to choose reasonable actions quickly. This problem is addressed by a number of action-selection mechanisms. Behaviour networks as proposed by Maes are one such mechanism, which is quite popular. In general, it seems not possible to predict when behaviour networks are well-behaved. However, they perform quite well in the robotic soccer context. In this paper, we analyse the reason for this success by identifying conditions that make behaviour networks goal converging, i.e., force them to reach the goals regardless of the details of the action selection scheme. In terms of STRIPS domains one could talk of self-solving planning domains.

Local deep feature learning framework for 3D shape. For 3D shape analysis, an effective and efficient feature is the key to popularize its applications in 3D domain. In this paper, we present a novel framework to learn and extract local deep feature (LDF), which encodes multiple low-level descriptors and provides high-discriminative representation of local region on 3D shape. The framework consists of four main steps. First, several basic descriptors are calculated and encapsulated to generate geometric bag-of-words in order to make full use of the various basic descriptors׳ properties. Then 3D mesh is down-sampled to hundreds of feature points for accelerating the model learning. Next, in order to preserve the local geometric information and establish the relationships among points in a local area, the geometric bag-of-words are encoded into local geodesic-aware bag-of-features (LGA-BoF). However, the resulting feature is redundant, which leads to low discriminative and efficiency. Therefore, in the final step, we use deep belief networks (DBNs) to learn a model, and use it to generate the LDF, which is high-discriminative and effective for 3D shape applications. 3D shape correspondence and symmetry detection experiments compared with related feature descriptors are carried out on several datasets and shape recognition is also conducted, validating the proposed local deep feature learning framework.

The complexity of belief update Abstract Belief revision and belief update are two different forms of belief change, and they serve dierent,purposes. In this paper we focus on belief update, the formalization of change in beliefs due to changes in the world. The complexity of the basic update (introduced by Winslett [1990]) has been determined in [Eiter and Gottlob, 1992]. Since then, many other formalizations have been proposed to overcome the limitations and drawbacks of Winslett’s update. In this paper we analyze the complexity of the proposals presented in the literature, and relate some of them to previous work on closed world reasoning.

Expressive Reasoning about Action in Nondeterministic Polynomial Time The rapid development of efficient heuristics for deciding satisfiability for propositional logic motivates thorough investigations of the usability of NP-complete problems in general. In this paper we introduce a logic of action and change which is expressive in the sense that it can represent most propositional benchmark examples in the literature, and some new examples involving parallel composition of actions, and actions that may or may not be executed. We prove that satisfiability of a scenario in this logic is NP-complete, and that it subsumes an NP-complete logic (which in turn includes a nontrivial polynomial-time fragment) previously introduced by Drakengren and Bjareland.

A Proposal for Describing Services with DLs Motivated by the semantic web application, we present a generic ex- tension of description logics to describe actions. These actions can then be chained to service descriptions. A web page providing a service can be annotated with a description of this service, which can then be taken into account by agents searching for a web service. Besides syntax and semantics of this extension of DLs, we dene and discuss inference prob- lems which are useful to annotate web pages with a description of the service they provide. 1 Motivation

Updates, actions, and planning A general framework for update-based planning is presented. We first give a new family of dependence-based update operators that are wellsuited to the representation of simple actions and we identify the complexity of query entailment from an updated belief base. Then we introduce conditional, nondeterministic and concurrent updates so as to encode the corresponding types of action effects. Plan verification and existence are expressed in this update-based framework.

From Causal Theories to Successor State Axioms and STRIPS-Like Systems We describe a system for specifying the efiects of actions. Unlike those commonly used in AI planning, our system uses an action description language that allows one to specify the efiects of actions using domain rules, which are state con- straints that can entail new action efiects from old ones. Declaratively, an action domain in our lan- guage corresponds to a nonmonotonic causal the- ory in the situation calculus. Procedurally, such an action domain is compiled into a set of proposi- tional theories, one for each action in the domain, from which fully instantiated successor state-like axioms and STRIPS-like systems are then gener- ated. We expect the system to be a useful tool for knowledge engineers writing action speciflca- tions for classical AI planning systems, GOLOG systems, and other systems where formal specifl- cations of actions are needed.

Possibilistic Planning: Representation and Complexity A possibilistic approach of planning under uncertainty has been developed recently. It applies to problems in which the initial state is partially known and the actions have graded nondeterministic effects, some being more possible (normal) than the others. The uncertainty on states and effects of actions is represented by possibility distributions. The paper first recalls the essence of possibilitic planning concerning the representational aspects and the plan generation algorithms used to...

Where "Ignoring delete lists" works: local search topology in planning benchmarks Between 1998 and 2004, the planning community has seen vast progress in terms of the sizes of benchmark examples that domain-independent planners can tackle successfully. The key technique behind this progress is the use of heuristic functions based on relaxing the planning task at hand, where the relaxation is to assume that all delete lists are empty. The unprecedented success of such methods, in many commonly used benchmark examples, calls for an understanding of what classes of domains these methods are well suited for. In the investigation at hand, we derive a formal background to such an understanding. We perform a case study covering a range of 30 commonly used STRIPS and ADL benchmark domains, including all examples used in the first four international planning competitions. We prove connections between domain structure and local search topology - heuristic cost surface properties - under an idealized version of the heuristic functions used in modern planners. The idealized heuristic function is called h+, and differs from the practically used functions in that it returns the length of an optimal relaxed plan, which is NP-hard to compute. We identify several key characteristics of the topology under h+, concerning the existence/non-existence of unrecognized dead ends, as well as the existence/non-existence of constant upper bounds on the difficulty of escaping local minima and benches. These distinctions divide the (set of all) planning domains into a taxonomy of classes of varying h+ topology. As it turns out, many of the 30 investigated domains lie in classes with a relatively easy topology. Most particularly, 12 of the domains lie in classes where FF's search algorithm, provided with h+, is a polynomial solving mechanism. We also present results relating h+ to its approximation as implemented in FF. The behavior regarding dead ends is provably the same. We summarize the results of an empirical investigation showing that, in many domains, the topological qualities of h+ are largely inherited by the approximation. The overall investigation gives a rare example of a successful analysis of the connections between typical-case problem structure, and search performance. The theoretical investigation also gives hints on how the topological phenomena might be automatically recognizable by domain analysis techniques. We outline some preliminary steps we made into that direction.

Representing action and change by logic programs We represent properties of actions in a logic programming language that uses both classical negation and negation as failure. The method is applicable to temporal projection problems with incomplete information, as well as to reasoning about the past. It is proved to be sound relative to a semantics of action based on states and transition functions.

Complexity, decidability and undecidability results for domain-independent planning In this paper, we examine how the complexity of domain-independent planning with STRIPS-style operators depends on the nature of the planning operators. We show conditions under which planning is decidable and undecidable. Our results on this topic solve an open problem posed by Chapman (5), and clear up some diculties with his undecidability theorems.

The complexity of Markov decision processes We investigate the complexity of the classical problem of optimal policy computation in Markov decision processes. All three variants of the problem finite horizon, infinite horizon discounted, and...

Equilibria and steering laws for planar formations This paper presents a Lie group setting for the problem of control of formations, as a natural outcome of the analysis of a planar two-vehicle formation control law. The vehicle trajectories are described using the planar Frenet–Serret equations of motion, which capture the evolution of both the vehicle position and orientation for unit-speed motion subject to curvature (steering) control. The set of all possible (relative) equilibria for arbitrary G-invariant curvature controls is described (where G=SE(2) is a symmetry group for the control law), and a global convergence result for the two-vehicle control law is proved. An n-vehicle generalization of the two-vehicle control law is also presented, and the corresponding (relative) equilibria for the n-vehicle problem are characterized. Work is on-going to discover stability and convergence results for the n-vehicle problem.

StagedDB: Designing Database Servers for Modern Hardware Advances in computer architecture research yield increasingly powerful processors which can execute code at a much faster pace than they can access data in the memory hierarchy. Database management systems (DBMS), due to their intensive data processing nature, are in the front line of commercial ap- plications which cannot harness the available computing power. To prevent the CPU from idling, a multitude of hardware mechanisms and software optimizations have been proposed. Their effectiveness, however, is limited by the sheer volume of data accessed and by the unpredictable sequence of memory requests. In this article we describe StagedDB, a new DBMS software architecture for optimizing data and instruction locality at all levels of the memory hierarchy. The key idea is to break database request execution in stages and process a group of sub-requests at each st age, thus effortlessly exploiting data and work commonality. We present two systems based on the StagedDB design. STEPS, a transac- tion coordinating mechanism demonstrated on top of Shore, minimizes instruction-cache misses without increasing the cache size, eliminating two thirds of all instruction misses when running on-line trans- action processing applications. QPipe, a staged relational query engine built on top of BerkeleyDB, maximizes data and work sharing across concurrent queries, providing up to 2x throughput speedup in a decision-support workload.

On the query complexity of clique size and maximum satisfiability This paper explores the bounded query complexity of approximating the size of the maximum clique in a graph (Clique Size) and the number of simultaneously satisfiable clauses in a 3CNF formula (MaxSat). The results in the paper show that for certain approximation factors, approximating Clique Size and MaxSat are complete for cor- responding bounded query classes under metric reductions. The completeness result is important because it shows that queries and approximation are interchangeable: NP queries can be used to solve NP-approximation problems and solutions to NP- approximation problems answer queries to NP oracles. Completeness also shows the existence of approximation preserving reductions from many NP-approximation prob- lems to approximating Clique Size and MaxSat (e.g., from approximating Chromatic Number to approximating Clique Size). Since query complexity is a quantitative com- plexity measure, these results also provide a framework for comparing the complexities of approximating Clique Size and approximating MaxSat. In addition, this paper ex- amines the query complexity of the minimization version of the satisfiability problem, MinUnsat, and shows that the complexity of approximating MinUnsat is very similar to the complexity of approximating Clique Size. Since MaxSat and MinUnsat share the same solution space, the "approximability" of MaxSat is not due to the intrinsic complexity of satisfiability, but is an artifact of viewing the approximation version of satisfiability as a maximization problem.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Security aspects in the cognition cycle of distributed cognitive radio networks: a survey from a multi-agent perspective Cognitive Radio CR enables Secondary Users SUs to observe, learn and take the right action on its operating environment at any time instant. This feature has inevitably exposed the SUs to artificial intelligence attacks from adversaries. This paper provides a survey on various kinds of attacks, as well as reviews and proposes mitigation approaches to tackle them from a multi-agent perspective. Additionally, this paper shows the impact and criticality of artificial intelligence attacks on network-wide performance. We aim to establish a foundation, and to spark new research interests in this new and critical area.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

How to achieve modularity in distributed object allocation The paper focuses on language constructs for driving the allocation of parallel object-oriented applications onto a target architecture. The paper analyses the issues that arise in the definition of these constructs and presents the solutions adopted in several systems and programming environments, by discussing their capability of enforcing the principle of modularity. Open issues and future directions of research are outlined.

Experiences on Porting a Parallel Objects Environment from a Transputer Network to a PVM-Based System Abstract: Parallel Objects is a powerful model for distributed/parallel Object-Oriented programming. Goal of this paper is to present the approach adopted in porting the support of the Parallel Objects environment, originally implemented for a massively parallel architecture, onto the PVM environment, which is nowadays a de-facto standard in the design of distributed applications on heterogeneous networks of computers.

High-level management of allocation in a parallel objects environment The diffusion of parallel architectures is currently limited by the lack of tools to exploit efficiently all available resources with few programming efforts. This is specially a problem in the allocation area. The paper presents a set of allocation tools (ACL) implemented within an object-oriented parallel programming environment. ACL defines a set of directives that permit users to specify the allocation needs of his/her applications without any knowledge of the architecture details. ACL directives drive the run-time support by tuning its general-purpose behaviour to the specific allocation needs of the applications. The effectiveness of the ACL approach is confirmed by testbed applications.

The Benefits Of Migration In A Parallel Objects Programming Environment.

The Narrowing Gap Between Language Systems and Operating Systems

Distributed transactions for reliable systems Facilities that support distributed transactions on user-defined types can be implemented efficiently and can simplify the construction of reliable distributed programs. To demonstrate these points, this paper describes a prototype transaction facility, called TABS, that supports objects, transparent communication, synchronization, recovery, and transaction management. Various objects that use the facilities of TABS are exemplified and the performance of the system is discussed in detail. The paper concludes that the prototype provides useful facilities, and that it would be feasible to build a high performance implementation based on its ideas.

Reliable object storage to support atomic actions Maintaining consistency of on-line, long-lived, distributed data in the presence of hardware failures is a necessity for many applications. The Argus programming language and system, currently under development at M.I.T., provides users with linguistic constructs to implement such applications. Argus permits users to identify certain data objects as being resilient to failures, and the set of such resilient objects can vary dynamically as programs run. When resilient objects are modified, they are automatically copied by the Argus implementation to stable storage, storage that with very high probability does not lose information. The resilient objects are therefore guaranteed, with very high probability, to survive both media failures and node crashes. This paper presents a method for implementing resilient objects, using a log-based mechanism to organize the information on stable storage. Of particular interest is the handling of a dynamic, user-controlled set of resilient objects, and the use of early prepare to minimize delays in user activities.

A combined method for maintaining large indices in multiprocessor multidisk environments Consider the problem of maintaining large indices (or secondary memory indices) in a multiprocessor multidisk environment in which each processor has a dedicated secondary memory (one disk or more). The processors either reside in the same site and communicate via shared memory, or reside in different sites and communicate via a local broadcast network. The straightforward method (SFM) for maintaining such an index, which is commonly called declustering, is to partition the index records equally among the processors, each of which maintains its part of the index in a local B/sup tree. In prior work (Inform. Processing Lett., vol. 34, pp. 313-321, May 1990), we have presented another method, called the "totally distributed B/sup tree" (TDB) method, in which all processors together implement a "wide" B/sup tree. There are settings in which the second method is better than the first method, and vice versa. In this paper, we present a new method, called the combined distribution method (CDM), that combines the ideas underlying SFM and TDB. In tightly coupled environments, CDM outperforms both SFM and TDB in almost all practical settings (in many settings by more than 30%). This is shown by an approximate analysis and verified by simulations. Note that CDM's approach can improve performance in database systems that use a RAID (redundant array of inexpensive disks).

Practical Issues in Temporal Difference Learning This paper examines whether temporal difference methods for training connectionist networks, such as Sutton's TD(λ) algorithm, can be successfully applied to complex real-world problems. A number of important practical issues are identified and discussed from a general theoretical perspective. These practical issues are then examined in the context of a case study in which TD(λ) is applied to learning the game of backgammon from the outcome of self-play. This is apparently the first application of this algorithm to a complex non-trivial task. It is found that, with zero knowledge built in, the network is able to learn from scratch to play the entire game at a fairly strong intermediate level of performance, which is clearly better than conventional commercial programs, and which in fact surpasses comparable networks trained on a massive human expert data set. This indicates that TD learning may work better in practice than one would expect based on current theory, and it suggests that further analysis of TD methods, as well as applications in other complex domains, may be worth investigating.

Global Continuation for Distance Geometry Problems Distance geometry problems arise in the determination of protein structure. We consider the case where only a subset of the distances between atoms is given and formulate this distance geometry problem as a global minimization problem with special structure. We show that global smoothing techniques and a continuation approach for global optimization can be used to determine global solutions of this problem reliably and efficiently. The global continuation approach determines a global solution with less computational effort than is required by a standard multistart algorithm. Moreover, the continuation approach usually finds the global solution from any given starting point, while the multistart algorithm tends to fail.

Recent Advances in AI Planning The past five years have seen dramatic advances in planning algorithms, with an emphasis on propositional methods such as Graphplan and compilers that convert planning problems into propositional CNF formulae for solution via systematic or stochastic SAT methods. Related work on the Deep Space One spacecraft control algorithms advances our understanding of interleaved planning and execution. In this survey,we explain the latest techniques and suggest areas for future research.

Context-aware prefetching at the storage server In many of today's applications, access to storage constitutes the major cost of processing a user request. Data prefetching has been used to alleviate the storage access latency. Under current prefetching techniques, the storage system prefetches a batch of blocks upon detecting an access pattern. However, the high level of concurrency in today's applications typically leads to interleaved block accesses, which makes detecting an access pattern a very challenging problem. Towards this, we propose and evaluate QuickMine, a novel, lightweight and minimally intrusive method for contextaware prefetching. Under QuickMine, we capture application contexts, such as a transaction or query, and leverage them for context-aware prediction and improved prefetching effectiveness in the storage cache. We implement a prototype of our context-aware prefetching algorithm in a storage-area network (SAN) built using Network Block Device (NBD). Our prototype shows that context-aware prefetching clearly out-performs existing context-oblivious prefetching algorithms, resulting in factors of up to 2 improvements in application latency for two e-commerce workloads with repeatable access patterns, TPC-W and RUBiS.

MAXSAT Heuristics for Cost Optimal Planning.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Latent Hierarchical Model of Temporal Structure for Complex Activity Classification Modeling the temporal structure of sub-activities is an important yet challenging problem in complex activity classification. This paper proposes a latent hierarchical model (LHM) to describe the decomposition of complex activity into sub-activities in a hierarchical way. The LHM has a tree-structure, where each node corresponds to a video segment (sub-activity) at certain temporal scale. The starting and ending time points of each sub-activity are represented by two latent variables, which are automatically determined during the inference process. We formulate the training problem of the LHM in a latent kernelized SVM framework and develop an efficient cascade inference method to speed up classification. The advantages of our methods come from: 1) LHM models the complex activity with a deep structure, which is decomposed into sub-activities in a coarse-to-fine manner and 2) the starting and ending time points of each segment are adaptively determined to deal with the temporal displacement and duration variation of sub-activity. We conduct experiments on three datasets: 1) the KTH; 2) the Hollywood2; and 3) the Olympic Sports. The experimental results show the effectiveness of the LHM in complex activity classification. With dense features, our LHM achieves the state-of-the-art performance on the Hollywood2 dataset and the Olympic Sports dataset.

On the expressive power of deep architectures Deep architectures are families of functions corresponding to deep circuits. Deep Learning algorithms are based on parametrizing such circuits and tuning their parameters so as to approximately optimize some training objective. Whereas it was thought too difficult to train deep architectures, several successful algorithms have been proposed in recent years. We review some of the theoretical motivations for deep architectures, as well as some of their practical successes, and propose directions of investigations to address some of the remaining challenges.

Learning long-term dependencies with gradient descent is difficult Recurrent neural networks can be used to map input sequences to output sequences, such as for recognition, production or prediction problems. However, practical difficulties have been reported in training recurrent neural networks to perform tasks in which the temporal contingencies present in the input/output sequences span long intervals. We show why gradient based learning algorithms face an increasingly difficult problem as the duration of the dependencies to be captured increases. These results expose a trade-off between efficient learning by gradient descent and latching on information for long periods. Based on an understanding of this problem, alternatives to standard gradient descent are considered.

Exploring Strategies for Training Deep Neural Networks Deep multi-layer neural networks have many levels of non-linearities allowing them to compactly represent highly non-linear and highly-varying functions. However, until recently it was not clear how to train such deep networks, since gradient-based optimization starting from random initialization often appears to get stuck in poor solutions. Hinton et al. recently proposed a greedy layer-wise unsupervised learning procedure relying on the training algorithm of restricted Boltzmann machines (RBM) to initialize the parameters of a deep belief network (DBN), a generative model with many layers of hidden causal variables. This was followed by the proposal of another greedy layer-wise procedure, relying on the usage of autoassociator networks. In the context of the above optimization problem, we study these algorithms empirically to better understand their success. Our experiments confirm the hypothesis that the greedy layer-wise unsupervised training strategy helps the optimization by initializing weights in a region near a good local minimum, but also implicitly acts as a sort of regularization that brings better generalization and encourages internal distributed representations that are high-level abstractions of the input. We also present a series of experiments aimed at evaluating the link between the performance of deep neural networks and practical aspects of their topology, for example, demonstrating cases where the addition of more depth helps. Finally, we empirically explore simple variants of these training algorithms, such as the use of different RBM input unit distributions, a simple way of combining gradient estimators to improve performance, as well as on-line versions of those algorithms.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Empirical Analysis of Predictive Algorithms for Collaborative Filtering Collaborative filtering or recommender systemsuse a database about user preferences topredict additional topics or products a newuser might like. In this paper we describe several algorithms designed for this task, including techniques based on correlation coefficients,vector-based similarity calculations,and statistical Bayesian methods. We comparethe predictive accuracy of the various methods in a set of representative problemdomains. We use two basic classes of evaluation...

Predicting individual disease risk based on medical history The monumental cost of health care, especially for chronic disease treatment, is quickly becoming unmanageable. This crisis has motivated the drive towards preventative medicine, where the primary concern is recognizing disease risk and taking action at the earliest signs. However, universal testing is neither time nor cost efficient. We propose CARE, a Collaborative Assessment and Recommendation Engine, which relies only on a patient's medical history using ICD-9-CM codes in order to predict future diseases risks. CARE uses collaborative filtering to predict each patient's greatest disease risks based on their own medical history and that of similar patients. We also describe an Iterative version, ICARE, which incorporates ensemble concepts for improved performance. These novel systems require no specialized information and provide predictions for medical conditions of all kinds in a single run. We present experimental results on a Medicare dataset, demonstrating that CARE and ICARE perform well at capturing future disease risks.

Real-time multimedia systems The expansion of multimedia networks and systems depends on real-time support for media streams and interactive multimedia services. Multimedia data are essentially continuous, heterogeneous, and isochronous, three characteristics with strong real-time implications when combined. At the same time, some multimedia services, like video-on-demand or distributed simulation, are real-time applications with sophisticated temporal functionalities in their user interface. We analyze the main problems in building such real-time multimedia systems, and we discuss-under an architectural prospect-some technological solutions especially those regarding determinism and efficient synchronization in the storage, processing, and communication of audio and video data

NP is as easy as detecting unique solutions For all known NP-complete problems the number of solutions in instances having solutions may vary over an exponentially large range. Furthermore, most of the well-known ones, such as satisfiability, are parsimoniously interreducible, and these can have any number of solutions between zero and an exponentially large number. It is natural to ask whether the inherent intractability of NP-complete problems is caused by this wide variation. In this paper we give a negative answer to this using randomized reductions. We show that the problems of distinguishing between instances of SAT having zero or one solution, or finding solutions to instances of SAT having unique solutions, are as hard as SAT itself. Several corollaries about the difficulty of specific problems follow. For example if the parity of the number of solutions of SAT can be computed in RP then NP = RP. Some further problems can be shown to be hard for NP or DP via randomized reductions.

DC++: distributed object-oriented system support on top of OSF DCE The OSF Distributed Computing Environment (DCE) is becoming an industry standard for open distributed computing. However, DCE only supports client/server-style applications based on the remote procedure call (RPC) communication model. This paper describes the design and imple- mentation of an extended distributed object-oriented environment, DC++, on top of DCE. As op- posed to RPC, it supports a uniform object model, location independent invocation of fine- grained objects, remote reference parameter passing, dynamic migration of objects between nodes, and C++ language integration. Moreover, the implementation is fully integrated with DCE, using DCE UUIDs for object identification, DCE threads for interobject concurrency, DCE RPC for remote object invocation, and the DCE Cell Directory Service (CDS) for optional re- trieval of objects by name. An additional stub compiler enables automatic generation of C++- based object communication interfaces. Low-level parameter encoding is done by DCE RPC's stub generation facility using the C-based DCE interface definition language (IDL). The system has been fully implemented and tested by implementing an office application. Experi- ences with the existing system and performance results are also reported in the paper. Further- more, a former, less transparent implementation of our group using DCE RPC as a pure transport- level mechanism is compared with the described approach. Related C++ extensions and stan- dardization efforts are also compared with our work.

Representing actions in logic programs and default theories a situation calculus approach We address the problem of representing common sense knowledge about action domains in the formalisms of logic programming and default logic. We employ a methodology proposed by Gelfond and Lifschitz which involves first defining a high-level language for representing knowledge about actions, and then specifying a translation from the high-level action language into a general-purpose formalism, such as logic programming. Accordingly, we define a high-level action languageAE, and specify sound and complete translations of portions ofAEinto logic programming and default logic. The languageAEincludes propositions that represent “static causal laws” of the following kind: a fluent formula ψ can be made true by making a fluent formula true (or, more precisely, ψ is caused whenever is caused). Such propositions are more expressive than the state constraints traditionally used to represent background knowledge. Our translations ofAEdomain descriptions into logic programming and default logic are simple, in part because the noncontrapositive nature of causal laws is easily reflected in such rule-based formalisms.

ARIMA time series modeling and forecasting for adaptive I/O prefetching Bursty application I/O patterns, together with transfer limited storage devices, combine to create a major I/O bottleneck on parallel systems. This paper explores the use of time series models to forecast application I/O request times, then prefetching I/O requests during computation intervals to hide I/O latency. Experimental results with I/O intensive scientific codes show performance improvements compared to standard UNIX prefetching strategies.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Profiler and compiler assisted adaptive I/O prefetching for shared storage caches I/O prefetching has been employed in the past as one of the mechanisms to hide large disk latencies. However, I/O prefetching in parallel applications is problematic when multiple CPUs share the same set of disks due to the possibility that prefetches from different CPUs can interact on shared memory caches in the I/O nodes in complex and unpredictable ways. In this paper, we (i) quantify the impact of compiler-directed I/O prefetching - developed originally in the context of sequential execution - on shared caches at I/O nodes. The experimental data collected shows that while I/O prefetching brings benefits, its effectiveness reduces significantly as the number of CPUs is increased; (ii) identify inter-CPU misses due to harmful prefetches as one of the main sources for this reduction in performance with the increased number of CPUs; and (iii) propose and experimentally evaluate a profiler and compiler assisted adaptive I/O prefetching scheme targeting shared storage caches. The proposed scheme obtains inter-thread data sharing information using profiling and, based on the captured data sharing patterns, divides the threads into clusters and assigns a separate (customized) I/O prefetcher thread for each cluster. In our approach, the compiler generates the I/O prefetching threads automatically. We implemented this new I/O prefetching scheme using a compiler and the PVFS file system running on Linux, and the empirical data collected clearly underline the importance of adapting I/O prefetching based on program phases. Specifically, our proposed scheme improves performance, on average, by 19.9%, 11.9% and 10.3% over the cases without I/O prefetching, with independent I/O prefetching (each CPU is performing compiler-directed I/O prefetching independently), and with one CPU prefetching (one CPU is reserved for prefetching on behalf of others), respectively, when 8 CPUs are used.

MAPFS: a flexible multiagent parallel file system for clusters The emergence of applications with greater processing and speedup requirements, such as Grand Challenge Applications (GCA), involves new computing and I/O needs. Many of these applications require access to huge data repositories and other I/O sources, making the I/O phase a bottleneck in the computing systems, due to its poor performance. In this sense, parallel I/O is becoming one of the major topics in the area of high-performance systems. Existing data-intensive GCA have been used in several domains, such as high energy physics, climate modeling, biology or visualization. Since the I/O problem has not been solved in this kind of applications, new approaches are required in this case. This paper presents MAPFS, a multiagent architecture, whose goal is to allow applications to access data in a cluster of workstations in an efficient and flexible fashion, providing formalisms for modifying the topology of the storage system, specifying different data access patterns and selecting additional functionalities.

A Decoupled Architecture for Application-Specific File Prefetching Data-intensive applications such as multimedia and data mining programs may exhibit sophisticated access patterns that are difficult to predict from past reference history and are different from one application to, another. This paper presents the design, implementation, and evaluation of an automatic application-specific file prefetching (AASFP) mechanism that is designed to improve the disk I/O performance of application programs with such complicated access patterns. The key idea of AASFP is to convert an application into two threads: a computation thread, which is the original program containing both computation and disk I/O, and a prefetch thread, which contains all the instructions in the original program that are related to disk accesses. At run time, the prefetch thread is scheduled to run sufficiently far ahead of the computation thread, so that disk blocks can be prefetched and put in the file buffer cache before the computation thread needs them. Through a source-to-source translator, the conversion of a given application into two such threads is made completely automatic. Measurements on an initial AASFP prototype under Linux show that it provides as much as 54% overall performance improvement for a volume visualization application.

Learning to classify parallel input/output access patterns Input/output performance on current parallel file systems is sensitive to a good match of application access patterns to file system capabilities. Automatic input/output access pattern classification can determine application access patterns at execution time, guiding adaptive file system policies. In this paper, we examine and compare two novel input/output access pattern classification methods based on learning algorithms. The first approach uses a feedforward neural network previously trained on access pattern benchmarks to generate qualitative classifications. The second approach uses hidden Markov models trained on access patterns from previous executions to create a probabilistic model of input/output accesses. In a parallel application, access patterns can be recognized at the level of each local thread or as the global interleaving of all application threads. Classification of patterns at both levels is important for parallel file system performance; we propose a method for forming global classifications from local classifications. We present results from parallel and sequential benchmarks and applications that demonstrate the viability of this approach.

Second-tier cache management using write hints Storage servers, as well as storage clients, typically have large memories in which they cache data blocks. This creates a two-tier cache hierarchy in which the presence of a first-tier cache (at the storage client) makes it more difficult to manage the second-tier cache (at the storage server). Many techniques have been proposed for improving the management of second-tier caches, but none of these techniques use the information that is provided by writes of data blocks from the first tier to help manage the second-tier cache. In this paper, we illustrate how the information contained in writes from the first tier can be used to improve the performance of the second-tier cache. In particular, we argue that there are very different reasons why storage clients write data blocks to storage servers (e.g., cleaning dirty blocks vs. limiting the time to recover from failure). These different types of writes can provide strong indications about the current state and future access patterns of a first-tier cache, which can help in managing the second-tier cache. We propose that storage clients inform the storage servers about the types of writes that they perform by passing write hints. These write hints can then be used by the server to manage the second-tier cache. We focus on the common and important case in which the storage client is a database system running a transactional (OLTP) workload. We describe, for this case, the different types of write hints that can be passed to the storage server, and we present several cache management policies that rely on these write hints. We demonstrate using trace driven simulations that these simple and inexpensive write hints can significantly improve the performance of the second-tier cache.

The Performance Impact of Kernel Prefetching on Buffer Cache Replacement Algorithms A fundamental challenge in improving file system performance is to design effective block replacement algorithms to minimize buffer cache misses. Despite the well-known interactions between prefetching and caching, almost all buffer cache replacement algorithms have been proposed and studied comparatively, without taking into account file system prefetching, which exists in all modern operating systems. This paper shows that such kernel prefetching can have a significant impact on the relative performance in terms of the number of actual disk I/Os of many well-known replacement algorithms; it can not only narrow the performance gap but also change the relative performance benefits of different algorithms. Moreover, since prefetching can increase the number of blocks clustered for each disk I/O and, hence, the time to complete the I/O, the reduction in the number of disk I/Os may not translate into proportional reduction in the total I/O time. These results demonstrate the importance of buffer caching research taking file system prefetching into consideration and comparing the actual disk I/Os and the execution time under different replacement algorithms.

Mining Sequential Patterns: Generalizations and Performance Improvements

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

A training algorithm for optimal margin classifiers A training algorithm that maximizes the margin between the training patterns and the decision boundary is presented. The technique is applicable to a wide variety of the classification functions, including Perceptrons, polynomials, and Radial Basis Functions. The effective number of parameters is adjusted automatically to match the complexity of the problem. The solution is expressed as a linear combination of supporting patterns. These are the subset of training patterns that are closest to the decision boundary. Bounds on the generalization performance based on the leave-one-out method and the VC-dimension are given. Experimental results on optical character recognition problems demonstrate the good generalization obtained when compared with other learning algorithms.

Representing actions: Laws, observations and hypotheses We propose a modificationL 1 of the action description languageA. The languageL 1 allows representation of hypothetical situations and hypothetical occurrence of actions (as inA) as well as representation of actual occurrences of actions and observations of the truth values of fluents in actual situations. The corresponding entailment relation formalizes various types of common-sense reasoning about actions and their effects not modeled by previous approaches. As an application of L1 we also present an architecture for intelligent agents capable of observing, planning and acting in a changing environment based on the entailment relation of L1 and use logic programming approximation of this entailment to implement a planning module for this architecture. We prove the soundness of our implementation and give a sufficient condition for its completeness.

Recent Advances in AI Planning The past five years have seen dramatic advances in planning algorithms, with an emphasis on propositional methods such as Graphplan and compilers that convert planning problems into propositional CNF formulae for solution via systematic or stochastic SAT methods. Related work on the Deep Space One spacecraft control algorithms advances our understanding of interleaved planning and execution. In this survey,we explain the latest techniques and suggest areas for future research.

Near-Optimal Parallel Prefetching and Caching Recently there has been a great deal of interest in the operating systems research community in prefetching and caching data from parallel disks, as a technique for enabling serial applications to improve input--output (I/O) performance. In this paper, algorithms are considered for integrated prefetching and caching in a model with a fixed-size cache and any number of backing storage devices (disks). The integration of caching and prefetching with a single disk was previously considered by Cao, Felten, Karlin, and Li. Here, it is shown that the natural extension of their aggressive algorithm to the parallel disk case is suboptimal by a factor near the number of disks in the worst case. The main result is a new algorithm, reverse aggressive, with near-optimal performance for integrated prefetching and caching in the presence of multiple disks.

MAXSAT Heuristics for Cost Optimal Planning.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

S/390 CMOS server I/O: The continuing evolution IBM has developed a strategy to achieve the high I/O demands of large servers. In a new environment of industry-standard peripheral component interconnect (PCI) attached adapters conforming to open I/O interfaces, S/390® has developed an efficient method of quickly integrating disk storage, communications, and future adapters. Preserving the S/390 I/O programming model and the high level of data integrity expected in S/390 products and reducing development cycle time and resources have further constrained design options. At the same time, S/390 developers have redesigned the traditional I/O components into the latest chip technologies. The developers have also designed a new internal link (STI) to meet the increased I/O bandwidth and connectivity required by the high processor performance of the third and fourth generations of S/390 CMOS servers. This paper describes this strategy and how it has led to systems that retain the differentiating features of S/390 products.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

Parameterized complexity for the database theorist

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Automated Control of Aggressive Prefetching for HTTP Streaming Video Servers Past work has shown that disk prefetching can be an effective technique for improving the performance of disk bound workloads. However, the performance gains are highly dependent on selecting a prefetch size that is appropriate for a specific system and workload. Using a prefetch size that is too small can lead to poor overall disk throughput, whereas prefetch sizes that are too large can lead to data being evicted before it can be used by a subsequent request. This paper looks at disk prefetch sizing for HTTP video streaming servers, such as those used by Apple, Adobe, Netflix, YouTube and Microsoft. We evaluate various representative streaming video workloads and show that the prefetch size that produces the best throughput can vary from 2 MB to 12 MB, and can depend on workload and system characteristics such as video bitrate, hard drive specifications, and memory capacity. A good choice of prefetch size can result in substantial performance gains, for example up to 3 times higher throughput than when using a prefetch size that is too large. We also find that application-level prefetching using the best prefetch size can provide up to 4 times higher throughput. In order to take full advantage of disk prefetching without extensive workload specific experimentation, we introduce an adaptive algorithm that dynamically selects an appropriate prefetch size. Most importantly, our results show our adaptive algorithm selects prefetch sizes that provide performance rivaling the best sizes determined through manual tuning, which requires extensive testing over different possible sizes.

Improve Prefetch Performance by Splitting the Cache Replacement Queue.

Using Libception to Understand and Improve HTTP Streaming Video Server Throughput. Video streaming applications generate a large fraction of Internet traffic. Much of this content is delivered over HTTP using standard web servers. Unlike other types of web workloads, HTTP video streaming workloads are typically disk bound, and therefore an important problem is that of optimizing disk access. In this paper we design, implement and evaluate Libception, an application-level shim library that implements techniques for improving disk I/O efficiency. Web servers can achieve the benefits of these techniques simply by linking with Libception, without the need to modify source code. In contrast to making kernel changes or attempting to optimize kernel tuning, Libception provides a portable and relatively simple setting in which techniques for optimizing I/O in HTTP video streaming servers can be implemented and evaluated. We report experimental results evaluating the efficacy of the aggressive prefetching and disk I/O serialization techniques currently implemented in Libception, for three web servers (Apache, nginx and the userver) and two operating systems (FreeBSD, Linux). We find that on FreeBSD, video streaming throughput with all three web servers can be doubled by linking with Libception. On Linux, performance similar to that provided with Libception was eventually obtained by examining the kernel source to understand and tune kernel parameters. With the default kernel parameter settings, however, and regardless of which Linux disk scheduler is selected, we find that use of Libception can approximately double throughput. We find that both aggressive prefetching and serialization are necessary to achieve these benefits.

A Prefetching Scheme Exploiting both Data Layout and Access History on Disk Prefetching is an important technique for improving effective hard disk performance. A prefetcher seeks to accurately predict which data will be requested and load it ahead of the arrival of the corresponding requests. Current disk prefetch policies in major operating systems track access patterns at the level of file abstraction. While this is useful for exploiting application-level access patterns, for two reasons file-level prefetching cannot realize the full performance improvements achievable by prefetching. First, certain prefetch opportunities can only be detected by knowing the data layout on disk, such as the contiguous layout of file metadata or data from multiple files. Second, nonsequential access of disk data (requiring disk head movement) is much slower than sequential access, and the performance penalty for mis-prefetching a randomly located block, relative to that of a sequential block, is correspondingly greater. To overcome the inherent limitations of prefetching at logical file level, we propose to perform prefetching directly at the level of disk layout, and in a portable way. Our technique, called DiskSeen, is intended to be supplementary to, and to work synergistically with, any present file-level prefetch policies. DiskSeen tracks the locations and access times of disk blocks and, based on analysis of their temporal and spatial relationships, seeks to improve the sequentiality of disk accesses and overall prefetching performance. It also implements a mechanism to minimize mis-prefetching, on a per-application basis, to mitigate the corresponding performance penalty. Our implementation of the DiskSeen scheme in the Linux 2.6 kernel shows that it can significantly improve the effectiveness of prefetching, reducing execution times by 20&percnt;--60&percnt; for microbenchmarks and real applications such as grep, CVS, and TPC-H. Even for workloads specifically designed to expose its weaknesses, DiskSeen incurs only minor performance loss.

Improving Disk Throughput in Data-Intensive Servers Low disk throughput is one of the main impediments to improving the performance of data-intensive servers. In this paper, we propose two management techniques for the disk controller cache that can significantly increase disk throughput. The first technique, called File-Oriented Read-ahead (FOR), adjusts the number of read-ahead blocks brought into the disk controller cache according to file system information. The second technique, called Host-guided Device Caching (HDC), gives the host control over part of the disk controller cache. As an example use of this mechanism, we keep the blocks that cause the most misses in the host buffer cache permanently cached in the disk controller. Our detailed simulations of real server workloads show that FOR and HDC can increase disk throughput by up to 34% and 24%, respectively, in comparison to conventional disk controller cache management techniques. When combined, the techniques can increase throughput by up to 47%.

Informed prefetching and caching The underutilization of disk parallelism and file cache buffers by traditional file systems induces I/O stall time that degrades the performance of modern microprocessor-based systems. In this paper, we present aggressive mechanisms that tailor file system resource management to the needs of I/O-intensive applications. In particular, we show how to use application-disclosed access patterns (hints) to expose and exploit I/O parallelism and to allocate dynamically file buffers among three competing demands: prefetching hinted blocks, caching hinted blocks for reuse, and caching recently used data for unhinted accesses. Our approach estimates the impact of alternative buffer allocations on application execution time and applies a cost-benefit analysis to allocate buffers where they will have the greatest impact. We implemented informed prefetching and caching in DEC''s OSF/1 operating system and measured its performance on a 150 MHz Alpha equipped with 15 disks running a range of applications including text search, 3D scientific visualization, relational database queries, speech recognition, and computational chemistry. Informed prefetching reduces the execution time of the first four of these applications by 20% to 87%. Informed caching reduces the execution time of the fifth application by up to 30%.

FS2: dynamic data replication in free disk space for improving disk performance and energy consumption Disk performance is increasingly limited by its head positioning latencies, i.e., seek time and rotational delay. To reduce the head positioning latencies, we propose a novel technique that dynamically places copies of data in file system's free blocks according to the disk access patterns observed at runtime. As one or more replicas can now be accessed in addition to their original data block, choosing the "nearest" replica that provides fastest access can significantly improve performance for disk I/O operations.We implemented and evaluated a prototype based on the popular Ext2 file system. In our prototype, since the file system layout is modified only by using the free/unused disk space (hence the name Free Space File System, or FS2), users are completely oblivious to how the file system layout is modified in the background; they will only notice performance improvements over time. For a wide range of workloads running under Linux, FS2 is shown to reduce disk access time by 41--68% (as a result of a 37--78% shorter seek time and a 31--68% shorter rotational delay) making a 16--34% overall user-perceived performance improvement. The reduced disk access time also leads to a 40--71% energy savings per access.

A trace-driven analysis of the UNIX 4.2 BSD file system

Disk cache—miss ratio analysis and design considerations The current trend of computer system technology is toward CPUs with rapidly increasing processing power and toward disk drives of rapidly increasing density, but with disk performance increasing very slowly if at all. The implication of these trends is that at some point the processing power of computer systems will be limited by the throughput of the input/output (I/O) system.A solution to this problem, which is described and evaluated in this paper, is disk cache. The idea is to buffer recently used portions of the disk address space in electronic storage. Empirically, it is shown that a large (e.g., 80-90 percent) fraction of all I/O requests are captured by a cache of an 8-Mbyte order-of-magnitude size for our workload sample. This paper considers a number of design parameters for such a cache (called cache disk or disk cache), including those that can be examined experimentally (cache location, cache size, migration algorithms, block sizes, etc.) and others (access time, bandwidth, multipathing, technology, consistency, error recovery, etc.) for which we have no relevant data or experiments. Consideration is given to both caches located in the I/O system, as with the storage controller, and those located in the CPU main memory. Experimental results are based on extensive trace-driven simulations using traces taken from three large IBM or IBM-compatible mainframe data processing installations. We find that disk cache is a powerful means of extending the performance limits of high-end computer systems.

Random Forests Random forests are a combination of tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the forest. The generalization error for forests converges a.s. to a limit as the number of trees in the forest becomes large. The generalization error of a forest of tree classifiers depends on the strength of the individual trees in the forest and the correlation between them. Using a random selection of features to split each node yields error rates that compare favorably to Adaboost (Y. Freund & R. Schapire, Machine Learning: Proceedings of the Thirteenth International conference, &ast;&ast;&ast;, 148–156), but are more robust with respect to noise. Internal estimates monitor error, strength, and correlation and these are used to show the response to increasing the number of features used in the splitting. Internal estimates are also used to measure variable importance. These ideas are also applicable to regression.

PatternHunter II: highly sensitive and fast homology search. Extending the single optimized spaced seed of PatternHunter to multiple ones, PatternHunter II simultaneously remedies the lack of sensitivity of Blastn and the lack of speed of Smith-Waterman, for homology search. At Blastn speed, PatternHunter II approaches Smith-Waterman sensitivity, bringing homology search technology back to a full circle.

Complexity of Data Tree Patterns over XML Documents We consider Boolean combinations of data tree patterns as a specification and query language for XML documents. Data tree patterns are tree patterns plus variable (in)equalities which express joins between attribute values. Data tree patterns are a simple and natural formalism for expressing properties of XML documents. We consider first the model checking problem (query evaluation), we show that it is DP-complete in general and already NP-complete when we consider a single pattern. We then consider the satisfiability problem in the presence of a DTD. We show that it is in general undecidable and we identify several decidable fragments.

Improving the tolerance of multilayer perceptrons by minimizing the statistical sensitivity to weight deviations This paper proposes a version of the backpropagation algorithm which increases the tolerance of a feedforward neural network against deviations in the weight values. These changes can originate either when the neural network is mapped on a given VLSI circuit where the precision and/or weight matching are low, or by physical defects affecting the neural circuits. The modified backpropagation algorithm we propose uses the statistical sensitivity of the network to changes in the weights as a quantitative measure of network tolerance and attempts to reduce this statistical sensitivity while keeping the figures for the usual training performance (in errors and time) similar to those obtained with the usual backpropagation algorithm.

Learning A Lexical Simplifier Using Wikipedia In this paper we introduce a new lexical simplification approach. We extract over 30K candidate lexical simplifications by identifying aligned words in a sentence-aligned corpus of English Wikipedia with Simple English Wikipedia. To apply these rules, we learn a feature-based ranker using SVMnk trained on a set of labeled simplifications collected using Amazon's Mechanical Turk. Using human simplifications for evaluation, we achieve a precision of 76% with changes in 86% of the examples.

Unsupervised Feature Learning using Self-organizing Maps.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Comparing Different Prenexing Strategies for Quantified Boolean Formulas The majority of the currently available solvers for quantified Boolean formulas (QBFs) process input formulas only in prenex conjunctive normal form. However, the natural representation of practicably relevant problems in terms of QBFs usually results in formulas which are not in a specific normal form. Hence, in order to evaluate such QBFs with available solvers, suitable normal-form translations are required. In this paper, we report experimental results comparing different prenexing strategies on a class of structured benchmark problems. The problems under consideration encode the evaluation of nested counterfactuals over a propositional knowledge base, and span the entire polynomial hierarchy. The results show that different prenexing strategies influence the evaluation time in different ways across different solvers. In particular, some solvers are robust to the chosen strategies while others are not.

Towards Implementations for Advanced Equivalence Checking in Answer-Set Programming In recent work, a general framework for specifying program corre- spondences under the answer-set semantics has been defined. The framework al- lows to define different notions of equivalence, including the well-known notions of strong and uniform equivalence, as well as refined equivalence notions based on the projection of answer sets, where not all parts of an answer set are of rel- evance (like, e.g., removal of auxiliary letters). In the general case, deciding the correspondence of two programs lies on the fourth level of the polynomial hierar- chy and therefore this task can (presumably) not be efficiently reduced to answer- set programming. In this paper, we describe an approach to compute program correspondences in this general framework by means of linear-time constructible reductions to quantified propositional logic. We can thus use extant solvers for the latter language as back-end inference engines for computing program corre- spondence problems. We also describe how our translations provide a method to construct counterexamples in case a program correspondence does not hold.

A Solver for QBFs in Nonprenex Form Various problems in artificial intelligence (AI) can be solved by translating them into a quantified boolean formula (QBF) and evaluating the resulting encoding. In this approach, a QBF solver is used as a black box in a rapid implementation of a more general reasoning system. Most of the current solvers for QBFs require formulas in prenex conjunctive normal form as input, which makes a further translation necessary, since the encodings are usually not in a specific normal form. This additional step increases the number of variables in the formula or disrupts the formula's structure. Moreover, the most important part of this transformation, prenexing, is not deterministic. In this paper, we focus on an alternative way to process QBFs without these drawbacks and describe a solver, qpro, which is able to handle arbitrary formulas. To this end, we extend algorithms for QBFs to the non-normal form case and compare qpro with the leading normal-form provers on problems from the area of AI.

Solving quantified boolean formulas with circuit observability don't cares Traditionally the propositional part of a Quantified Boolean Formula (QBF) instance has been represented using a conjunctive normal form (CNF). As with propositional satisfiability (SAT), this is motivated by the efficiency of this data structure. However, in many cases, part of or the entire propositional part of a QBF instance can often be represented as a combinational logic circuit. In a logic circuit, the limited observability of the internal signals at the circuit outputs may make their assignments irrelevant for specific assignments of values to other signals in the circuit. This circuit observability don't care (ODC) information has been used to advantage in circuit based SAT solvers. A CNF encoding of the circuit, however, does not capture the signal direction and this limited observability, and thus cannot directly take advantage of this. However, recently it has been shown that this don't care information can be encoded in the CNF description and taken advantage of in a DPLL based SAT solver by modifying the decision heuristics/Boolean constraint propagation/conflict-driven-learning to account for these don't cares. Thus far, however, the use of these don't cares in the CNF encoding has not been explored for QBF solvers. In this paper, we examine how this can be done for QBF solvers as well as evaluate its practical benefits through experimentation. We have developed and implemented the usage of circuit ODCs in various parts of the DPLL-based procedure of the Quaffle QBF solver. We show that DPLL search based QBF solvers can use circuit ODC information to detect satisfying branches earlier during search and make satisfiability directed learning more effective. Our experiments demonstrate that significant performance gain can be obtained by considering circuit ODCs in checking the satisfiability of QBFs.

On Computing Belief Change Operations using Quantified Boolean Formulas In this paper, we show how an approach to belief revision and belief contraction can be axiomatized by means of quantified Boolean formulas. Specifically, we consider the approach of belief change scenarios, a general framework that has been introduced for expressing different forms of belief change. The essential idea is that for a belief change scenario (K, R, C), the set of formulas K, representing the knowledge base, is modified so that the sets of formulas R and C are respectively true in, and consistent with the result. By restricting the form of a belief change scenario, one obtains specific belief change operators including belief revision, contraction, update, and merging. For both the general approach and for specific operators, we give a quantified Boolean formula such that satisfying truth assignments to the free variables correspond to belief change extensions in the original approach. Hence, we reduce the problem of determining the results of a belief change operation to that of satisfiability. This approach has several benefits. First, it furnishes an axiomatic specification of belief change with respect to belief change scenarios. This then leads to further insight into the belief change framework. Second, this axiomatization allows us to identify strict complexity bounds for the considered reasoning tasks. Third, we have implemented these different forms of belief change by means of existing solvers for quantified Boolean formulas. As well, it appears that this approach may be straightforwardly applied to other specific approaches to belief change.

Backjumping for quantified Boolean logic satisfiability The implementation of effective reasoning tools for deciding the satisfiability of Quantified Boolean Formulas (QBFs) is an important research issue in Artificial Intelligence. Many decision procedures have been proposed in the last few years, most of them based on the Davis, Logemann, Loveland procedure (DLL) for propositional satisfiability (SAT). In this paper we show how it is possible to extend the conflict-directed backjumping schema for SAT to the satisfiability of QBFs: When applicable, conflict-directed backjumping allows search to skip over existentially quantified literals while backtracking. We introduce solution-directed backjumping, which allows the same behavior for universally quantified literals. We show how it is possible to incorporate both conflict-directed and solution-directed backjumping in a DLL-based decision procedure for satisfiability of QBFs. We also implement and test the procedure: The experimental analysis shows that, because of backjumping, significant speed-ups can be obtained.Summing up: We present the first algorithm that applies conflict and solution directed backjumping to QBF, and demonstrate the performance of this algorithm via an empirical study.

An algorithm to evaluate quantified Boolean formulae The high computational complexity of advanced reasoning tasks such as belief revision and planning calls for efficient and reliable algorithms for reasoning problems harder than NP. In this paper we propose Evaluate, an algorithm for evaluating Quantified Boolean Formulae, a language that extends propositional logic in a way such that many advanced forms of propositional reasoning, e.g., reasoning about knowledge, can be easily formulated as evaluation of a QBF. Algorithms for evaluation of QBFs are suitable for the experimental analysis on a wide range of complexity classes, a property not easily found in other formalisms. Evaluate is based on a generalization of the Davis-Putnam procedure for SAT, and is guaranteed to work in polynomial space. Before presenting Evaluate, we discuss all the abstract properties of QBFs that we singled out to make the algorithm more efficient. We also briefly mention the main results of the experimental analysis, which is reported elsewhere.

QUIP - A Tool for Computing Nonmonotonic Reasoning Tasks In this paper, we outline the prototype of an auto- mated inference tool, called QUIP, which provides a uniform implementation for several nonmonotonic rea- soning formalisms. The theoretical basis of QUIP is de- rived from well-known results about the computational complexity of nonmonotonic logics and exploits a rep- resentation of the different reasoning tasks in terms of quantified boolean formulae (QBFs).

Abductive reasoning through filtering Abduction is an inference mechanism where given a knowledge base and some observations,the reasoner tries to find hypotheses which together with the knowledgebase explain the observations. A reasoning based on such an inference mechanismis referred to as abductive reasoning. Given a theory and some observations, by filtering the theory with the observations, we mean selecting only those models of thetheory that entail the observations. Entailment with respect to these selected modelsis...

A causal theory of ramifications and qualifications This paper is concerned with the problem of determining the indirect effects or ramifications of actions. We argue that the standard framework in which background knowledge is given in the form of state constraints is inadequate and that background knowledge should instead be given in the form of "causal laws." We represent "causal laws" first as inference rules and later as sentences in a modal, conditional logic Gflat- For the framework with "causal laws," we propose a simple fixpoint condition defining the possible next states after performing an action. This fixpoint condition guarantees minimal change between states, but also enforces the requirement that changes be "caused." Ramification and qualification constraints can be expressed as "causal laws."

Sequentiality and prefetching in database systems Sequentiality of access is an inherent characteristic of many database systems. We use this observation to develop an algorithm which selectively prefetches data blocks ahead of the point of reference. The number of blocks prefetched is chosen by using the empirical run length distribution and conditioning on the observed number of sequential block references immediately preceding reference to the current block. The optimal number of blocks to prefetch is estimated as a function of a number of “costs,” including the cost of accessing a block not resident in the buffer (a miss), the cost of fetching additional data blocks at fault times, and the cost of fetching blocks that are never referenced. We estimate this latter cost, described as memory pollution, in two ways. We consider the treatment (in the replacement algorithm) of prefetched blocks, whether they are treated as referenced or not, and find that it makes very little difference. Trace data taken from an operational IMS database system is analyzed and the results are presented. We show how to determine optimal block sizes. We find that anticipatory fetching of data can lead to significant improvements in system operation.

Automatic Parallel I/O Performance Optimization Using Genetic Algorithms The complexity of parallel I/O systems imposes significant challenge in managing and utilizing the available system resources to meet application performance, portability and usability goals. We believe that a parallel I/O system that automatically selects efficient I/O plans for user applications is a solution to this problem. In this paper, we present such an automatic performance optimization approach for scientific applications performing collective I/O requests on multidimensional arrays. The approach is based on a high level description of the target workload and execution environment characteristics, and applies genetic algorithms to select high quality I/O plans. We have validated this approach in the Panda parallel I/O library. Our performance evaluations on the IBM SP show that this approach can select high quality I/O plans under a variety of system conditions with a low overhead, and the genetic algorithm-selected I/O plans are in general better than the default plans used in Panda.

Introduction: progress in formal commonsense reasoning This special issue consists largely of expanded and revised versions of selected papers of the Fifth International Symposium on Logical Formalizations of Commonsense Reasoning (Common Sense 2001), held at New York University in May 2001., The Common Sense Symposia, first organized in 1991 by John McCarthy and held roughly biannually since, are dedicated to exploring the development of formal commonsense theories using mathematical logic. Commonsense reasoning is a central part of human behavior; no real intelligence is possible without it. Thus, the development of systems that exhibit commonsense behavior is a central goal of Artificial Intelligence. It has proven to be more difficult to create systems that are capable of commonsense reasoning than systems that can solve "hard" reasoning problems. There are chess-playing programs that beat champions [5] and expert systems that assist in clinical diagnosis [32], but no programs that reason about how far one must bend over to put on one's socks. Part of the difficulty is the all-encompassing aspect of commonsense reasoning: any problem one looks at touches on many different types of knowledge. Moreover, in contrast to expert knowledge which is usually explicit, most commonsense knowledge is implicit. One of the prerequisites to developing commonsense reasoning systems is making this knowledge explicit. John McCarthy [25] first noted this need and suggested using formal logic to encode commonsense knowledge and reasoning. In the ensuing decades, there has been much research on the representation of knowledge in formal logic and on inference algorithms to

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

A machine program for theorem-proving The programming of a proof procedure is discussed in connection with trial runs and possible improvements.

Unified QBF certification and its applications Quantified Boolean formulae (QBF) allow compact encoding of many decision problems. Their importance motivated the development of fast QBF solvers. Certifying the results of a QBF solver not only ensures correctness, but also enables certain synthesis and verification tasks. To date the certificate of a true formula can be in the form of either a syntactic cube-resolution proof or a semantic Skolem-function model whereas that of a false formula is only in the form of a syntactic clause-resolution proof. The semantic certificate for a false QBF is missing, and the syntactic and semantic certificates are somewhat unrelated. This paper identifies the missing Herbrand-function countermodel for false QBF, and strengthens the connection between syntactic and semantic certificates by showing that, given a true QBF, its Skolem-function model is derivable from its cube-resolution proof of satisfiability as well as from its clause-resolution proof of unsatisfiability under formula negation. Consequently Skolem-function derivation can be decoupled from special Skolemization-based solvers and computed from standard search-based ones. Experimental results show strong benefits of the new method.

Verification of partial designs using incremental QBF. SAT solving is an indispensable core component of numerous formal verification tools and has found widespread use in industry, in particular when using it in an incremental fashion, e.g., in Bounded Model Checking (BMC). On the other hand, for some applications SAT formulas are not expressive enough, whereas a description via Quantified Boolean Formulas (QBF) is much more adequate, for instance when dealing with partial designs.Motivated by the success of incremental SAT, in this paper we explore various approaches to solve QBF problems in an incremental fashion and thereby make this technology usable as a core component of BMC. Firstly, we realized an incremental QBF solver based on the state-of-the-art QBF solver QuBE: Taking profit from the reuse of some information from previous iterations, the search space can be pruned, in some cases, to even less than a quarter.However, the need for preprocessing QBF formulas prior to the solving phase, that in general cannot be paired with incremental solving because of the non-predictable elimination of variables in the future incremental steps, posed the question of incremental QBF preprocessing. In this context we present an approach for retaining the QBF formula being preprocessed while extending its clauses and prefix incrementally. This procedure results in a significant size reduction of the QBF formulas, hence leading to a reduced solving time.As this may come together with a high preprocessing time, we analyze various heuristics to dynamically disable incremental preprocessing when its overhead raises over a certain threshold and is not compensated by the reduced solving time anymore.For proving the efficacy of our methods experimentally, as an application we consider BMC for partial designs (i.e., designs containing so-called blackboxes which represent unknown parts). Here, we disprove realizability, that is, we prove that an unsafe state is reachable no matter how the blackboxes are implemented. We examine all these incremental approaches from both the point of view of the effectiveness of the single procedure and the benefits that a range of QBF solvers can take from it. On a domain of partial design benchmarks, engaging incremental QBF methods significant performance gains over non incremental BMC can be achieved.

Conformant Planning as a Case Study of Incremental QBF Solving. We consider planning with uncertainty in the initial state as a case study of incremental quantified Boolean formula (QBF) solving. We report on experiments with a workflow to incrementally encode a planning instance into a sequence of QBFs. To solve this sequence of successively constructed QBFs, we use our general-purpose incremental QBF solver DepQBF. Since the generated QBFs have many clauses and variables in common, our approach avoids redundancy both in the encoding phase as well as in the solving phase. We also present experiments with incremental preprocessing techniques that are based on blocked clause elimination (QBCE). QBCE allows to eliminate certain clauses from a QBF in a satisfiability preserving way. We implemented the QBCE-based techniques in DepQBF in three variants: as preprocessing, as inprocessing (which extends preprocessing by taking into account variable assignments that were fixed by the QBF solver), and as a novel dynamic approach where QBCE is tightly integrated in the solving process. For DepQBF, experimental results show that incremental QBF solving with incremental QBCE outperforms incremental QBF solving without QBCE, which in turn outperforms nonincremental QBF solving. For the first time we report on incremental QBF solving with incremental QBCE as inprocessing. Our results are the first empirical study of incremental QBF solving in the context of planning and motivate its use in other application domains.

QuBE++: An Efficient QBF Solver In this paper we describe QuBE++, an efficient solver for Quantified Boolean Formulas (QBFs). To the extent of our knowledge, QUBE++ is the first QBF reasoning engine that uses lazy data structures both for unit clauses propagation and for pure literals detection. QuBE++ also features non-chronological backtracking and a branching heuristic that leverages the information gathered during the backtracking phase. Owing to such techniques and to a careful implementation, QuBE++ turns out to be an efficient and robust solver, whose performances exceed those of other state-of-the-art QBF engines, and are comparable with the best engines currently available on SAT instances.

The good, the bad, and the odd: cycles in answer-set programs Backdoors of answer-set programs are sets of atoms that represent "clever reasoning shortcuts" through the search space. Assignments to backdoor atoms reduce the given program to several programs that belong to a tractable target class. Previous research has considered target classes based on notions of acyclicity where various types of cycles (good and bad cycles) are excluded from graph representations of programs. We generalize the target classes by taking the parity of the number of negative edges on bad cycles into account and consider backdoors for such classes. We establish new hardness results and non-uniform polynomial-time tractability relative to directed or undirected cycles.

Learning for quantified boolean logic satisfiability Learning, i.e., the ability to record and exploit some information which is unveiled during the search, proved to be a very effective AI technique for problem solving and, in particular, for constraint satisfaction. We introduce learning as a general purpose technique to improve the performances of decision procedures for Quantified Boolean Formulas (QBFs). Since many of the recently proposed decision procedures for QBFs solve the formula using search methods, the addition of learning to such procedures has the potential of reducing useless explorations of the search space. To show the applicability of learning for QBF satisfiability we have implemented it in QUBE, a state-of-the-art QBF solver. While the backjumping engine embedded in QUBE provides a good starting point for our task, the addition of learning required us to devise new data structures and led to the definition and implementation of new pruning strategies. We report some experimental results that witness the effectiveness of learning. Noticeably, QUBE augmented with learning is able to solve instances that were previously out if its reach. To the extent of our knowledge, this is the first time that learning is proposed, implemented and tested for QBFs satisfiability.

An Effective QBF Solver for Planning Problems A large number of applications can be represented by quantified Boolean formulas (QBF). Although evaluating QBF is NP-hard and thus very difficult, there has been significant progress in the development of QBF solvers. These solvers require the quantified Boolean formula to be in a standard format. We have encountered a large class of problems whose representation as QBF is not in that standard format. If we apply current state-of-the-art QBF solvers, the required transformation into standard format increases the size of the formula and tends to hide structural properties of the problem class. We suggest a direct attack of the problem. The solution algorithm is based on backtracking search and on a new form of learning clauses. We have tested a first implementation of the algorithm on a class of planning problems. The tests show that the approach is significantly faster than current state-of-the-art QBF solvers.

Approximation of action theories and its application to conformant planning This paper describes our methodology for building conformant planners, which is based on recent advances in the theory of action and change and answer set programming. The development of a planner for a given dynamic domain starts with encoding the knowledge about fluents and actions of the domain as an action theory D of some action language. Our choice in this paper is AL - an action language with dynamic and static causal laws and executability conditions. An action theory D of AL defines a transition diagram T(D) containing all the possible trajectories of the domain. A transition belongs to T(D) iff the execution of the action a in the state s may move the domain to the state s^'. The second step in the planner development consists in finding a deterministic transition diagram T^l^p(D) such that nodes of T^l^p(D) are partial states of D, its arcs are labeled by actions, and a path in T^l^p(D) from an initial partial state @d^0 to a partial state satisfying the goal @d^f corresponds to a conformant plan for @d^0 and @d^f in T(D). The transition diagram T^l^p(D) is called an 'approximation' of T(D). We claim that a concise description of an approximation of T(D) can often be given by a logic program @p(D) under the answer sets semantics. Moreover, complex initial situations and constraints on plans can be also expressed by logic programming rules and included in @p(D). If this is possible then the problem of finding a parallel or sequential conformant plan can be reduced to computing answer sets of @p(D). This can be done by general purpose answer set solvers. If plans are sequential and long then this method can be too time consuming. In this case, @p(D) is used as a specification for a procedural graph searching conformant planning algorithm. The paper illustrates this methodology by building several conformant planners which work for domains with complex relationship between the fluents. The efficiency of the planners is experimentally evaluated on a number of new and old benchmarks. In addition we show that for a subclass of action theories of AL our planners are complete, i.e., if in T^l^p(D) we cannot get from @d^0 to a state satisfying the goal @d^f then there is no conformant plan for @d^0 and @d^f in T(D).

Automatic Memory Reductions for RTL Model Verification We present several techniques for automatically reducing memories in RTL designs. This includes a new memory abstraction algorithm that allows us to greatly reduce the size of memories and a technique based on-term rewriting that further improves the abstraction. In contrast to previously proposed methods for abstracting memories of RTL designs, our methods are general---e.g., they allow us to arbitrarily and directly compare memories---and they are sound and complete---e.g., there are no false positives or negatives. In addition, the combination of our techniques allows us to automatically verify RTL pipelined machine designs beyond the reach of current state-of-the-art methods, as our experimental results show.

Two components of an action language Some of the recent work on representing action makes use of high&dash;level action languages. In this paper we show that an action language can be represented as the sum of two distinct parts: an “action description language” and an “action query language.” A set of propositions in an action description language describes the effects of actions on states. Mathematically, it defines a transition system of the kind familiar from the theory of finite automata. An action query language serves for expressing properties of paths in a given transition system. We define the general concepts of a transition system, of an action description language and of an action query language, give a series of examples of languages of both kinds, and show how to combine a description language and a query language into one. This construction makes it possible to design the two components of an action language independently, which leads to the simplification and clarification of the theory of actions.

Generalization by weight-elimination with application to forecasting Inspired by the information theoretic idea of minimum description length, we add a term to the back propagation cost function that penalizes network complexity. We give the details of the procedure, called weight-elimination, describe its dynamics, and clarify the meaning of the parameters involved. From a Bayesian perspective, the complexity term can be usefully interpreted as an assumption about prior distribution of the weights. We use this procedure to predict the sunspot time series and the notoriously noisy series of currency exchange rates.

On Complexity of Counting Let l, u: , l. We give a full characterization of intervals [l, u] such that a polynomial-time ATM of a constant numer of alternations can verify the number of words of a given length and in a given (as its oracle) set A, provided that A's density function is in [l, u]. We prove also a new lower bound on the approximate counting: there is a recursive set A whose elements cannot be approximate counted in 2 p, A 2 p, A .

Improving Citation Polarity Classification With Product Reviews Recent work classifying citations in scientific literature has shown that it is possible to improve classification results with extensive feature engineering. While this result confirms that citation classification is feasible, there are two drawbacks to this approach: (i) it requires a large annotated corpus for supervised classification, which in the case of scientific literature is quite expensive; and (ii) feature engineering that is too specific to one area of scientific literature may not be portable to other domains, even within scientific literature. In this paper we address these two drawbacks. First, we frame citation classification as a domain adaptation task and leverage the abundant labeled data available in other domains. Then, to avoid over-engineering specific citation features for a particular scientific domain, we explore a deep learning neural network approach that has shown to generalize well across domains using unigram and bigram features. We achieve better citation classification results with this cross-domain approach than using in-domain classification.

A fast BP networks with dynamic sample selection for handwritten recognition. Training time of traditional multilayer perceptrons (MLPs) using back-propagation algorithm rises seriously with the problem scale. For multi-class problems, the convergence ratio is very low for training MLPs. The huge time-consuming and low convergence ratio greatly restricts the applications of MLPs on problems with tens and thousands of samples. To deal with these disadvantages, this paper proposes a fast BP network with dynamic sample selection (BPNDSS) method which can dynamically select the samples containing more contribution to the variation of the decision boundary for training after each iteration epoch. The proposed BPNDSS can significantly increase the training speed by only selecting a small subset of the whole samples. Moreover, two kinds of modular single-hidden-layer approaches are adopted to decompose a multi-class problem into multiple binary-class sub-problems, which result in the high rate of convergence. The experiments on Letter and MNIST handwritten recognition database show the effectiveness and the efficiency of BPNDSS. Moreover, BPNDSS results in comparable classification performance to the convolutional neural networks (CNNs), support vector machine, Adaboost, C4.5, and nearest neighbour algorithms. To further demonstrate the training speed improvement of the dynamic sample selection approach on large-scale datasets, we modify CNN to propose a dynamic sample selection CNN (DynCNN). Experiments on Image-Net dataset illustrate that DynCNN can result in similar performance to CNN, but consume less training time.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

Load-balancing for WAN warehouses Although the basic Data Warehouse schema concept is centralized, there are increasingly application domains in which there is the need to have several sites or computers input and analyze the data, therefore distributed data placement and processing is necessary. Given that sites may have different amounts of data and different processing capacities, how can we conform to the placement requirements of the context and balance such a system effectively? In WAN environments the network speed is a very relevant factor and there are application requirements concerning the place where each piece of data stays, based on who produced the data (ownership). We propose a new strategy that accepts the placement requirements of the desired context and uses an effective automatic approach to determine fixed-sized chunks and to balance and process those chunks efficiently. Our experimental results show the validity of the approach and how to minimize the context limitations.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Subgraph-augmented Path Embedding for Semantic User Search on Heterogeneous Social Network. Semantic user search is an important task on heterogeneous social networks. Its core problem is to measure the proximity between two user objects in the network w.r.t. certain semantic user relation. State-of-the-art solutions often take a path-based approach, which uses the sequences of objects connecting a query user and a target user to measure their proximity. Despite their success, we assert that path as a low-order structure is insufficient to capture the rich semantics between two users. Therefore, in this paper we introduce a new concept of subgraph-augmented path for semantic user search. Specifically, we consider sampling a set of object paths from a query user to a target user; then in each object path, we replace the linear object sequence between its every two neighboring users with their shared subgraph instances. Such subgraph-augmented paths are expected to leverage both path»s distance awareness and subgraph»s high-order structure. As it is non-trivial to model such subgraph-augmented paths, we develop a Subgraph-augmented Path Embedding (SPE) framework to accomplish the task. We evaluate our solution on six semantic user relations in three real-world public data sets, and show that it outperforms the baselines.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Learning Topic Representation For Smt With Neural Networks Statistical Machine Translation (SMT) usually utilizes contextual information to disambiguate translation candidates. However, it is often limited to contexts within sentence boundaries, hence broader topical information cannot be leveraged. In this paper, we propose a novel approach to learning topic representation for parallel data using a neural network architecture, where abundant topical contexts are embedded via topic relevant monolingual data. By associating each translation rule with the topic representation, topic relevant rules are selected according to the distributional similarity with the source text during SMT decoding. Experimental results show that our method significantly improves translation accuracy in the NIST Chinese-to-English translation task compared to a state-of-the-art baseline.

HOPPER: a hierarchical planning agent for unpredictable domains Hierarchical Task Networks (HTNs) are a family of powerful planning algorithms that have been successfully applied to many complex, real-world domains. However, they are limited to predictable domains. In this paper we present HOPPER (Hierarchical Ordered Partial-Plan Executor and Re-planner), a hierarchical planning agent that produces partial plans in a similar way to HTNs but can also handle unexpected events in unpredictable domains by interleaving planning and execution. HOPPER can detect and recover from unexpected events that invalidate the plan, and it can detect and exploit unexpected opportunities both serendipitously and by interleaving decompositions.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

Parameterized complexity for the database theorist

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.

Learning Polynomials with Neural Networks.

A logic for default reasoning The need to make default assumptions is frequently encountered in reasoning'about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non.monotonJcity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occurring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.

Extended stable semantics for normal and disjunctive programs

The Stable Model Semantics for Logic Programming We propose a new declarative semantices for logic programs with negation.Its formulation is quite simple;at the same time, it is more general than the iterated fixed point semantics for stratified programs,and is applicable to some useful programs that are not stratified.

Classical Negation in Logic Programs and Disjunctive Databases An important limitation of traditional logic programming as a knowledge representation tool, in comparison with classical logic, is that logic programming does not allow us to deal directly with incomplete information. In order to overcome this limitation, we extend the class of general logic programs by including classical negation, in addition to negation-as-failure. The semantics of such extended programs is based on the method of stable models. The concept of a disjunctive database can be extended in a similar way. We show that some facts of commonsense knowledge can be represented by logic programs and disjunctive databases more easily when classical negation is available.Computationally, classical negation can be eliminated from extended programs by a simple preprocessor. Extended programs are identical to a special case of default theories in the sense of Reiter.

Improvements to the Evaluation of Quantified Boolean Formulae We present a theorem-prover for quantified Boolean formulae and evaluate it on random quantified formulae and formulae that represent problems from automated planning. Even though the notion of quantified Boolean formula is theoretically important, automated reasoning with QBF has not been thoroughly investigated. Universal quantifiers are needed in representing many computational problems that cannot be easily translated to the propositional logic and solved by satisfiability algorithms. Therefore efficient reasoning with QBF is important. The Davis-Putnam procedure can be extended to evaluate quantified Boolean formulae. A straightforward algorithm of this kind is not very efficient. We identify universal quantifiers as the main area where improvements to the basic algorithm can be made. We present a number of techniques for reducing the amount of search that is needed, and evaluate their effectiveness by running the algorithm on a collection of formulae obtained from planning and generated randomly. For the structured problems we consider, the techniques lead to a dramatic speed-up.

Multi-level transaction management for complex objects: implementation, performance, parallelism Multi-level transactions are a variant of open-nested transactions in which the subtransactions correspond to operations at different levels of a layered system architecture. They allow the exploitation of semantics of high-level operations to increase concurrency. As a consequence, undoing a transaction requires compensation of completed subtransactions. In addition, multi-level recovery methods must take into consideration that high-level operations are not necessarily atomic if multiple pages are updated in a single subtransaction. This article presents algorithms for multi-level transaction management that are implemented in the database kernel system (DASDBS). In particular, we show that multi-level recovery can be implemented in an efficient way. We discuss performance measurements using a synthetic benchmark for processing complex objects in a multi-user environment. We show that multi-level transaction management can be extended easily to cope with parallel subtransactions within a single transaction. Performance results are presented with varying degrees of inter- and intratransaction parallelism.

Nonlinear component analysis as a kernel eigenvalue problem A new method for performing a nonlinear form of principal component analysis is proposed. By the use of integral operator kernel functions, one can efficiently compute principal components in high-dimensional feature spaces, related to input space by some nonlinear map-for instance, the space of all possible five-pixel products in 16 x 16 images. We give the derivation of the method and present experimental results on polynomial feature extraction for pattern recognition.

Performance of a mirrored disk in a real-time transaction system Disk mirroring has found widespread use in computer systems as a method for providing fault tolerance. In addition to increasing reliability, a mirrored disk can also reduce I/O response time by supporting the execution of parallel I/O requests. The improvement in I/O efficiency is extremely important in a real-time system, where each computational entity carries a deadline. In this paper, we present two classes of real-time disk scheduling policies, RT-DMQ and RT-CMQ, for a mirrored disk I/O subsystem and examine their performance in an integrated real-time transaction system. The real-time transaction system model is validated on a real-time database testbed, called RT-CARAT. The performance results show that a mirrored disk I/O subsystem can decrease the fraction of transactions that miss their deadlines over a single disk system by 68%. Our results also reveal the importance of real-time scheduling policies, which can lead up to a 17% performance improvement over non-real-time policies in terms of minimizing the transaction loss ratio.

Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing Solving the SLAM (simultaneous localization and mapping) problem is one way to enable a robot to explore, map, and navigate in a previously unknown environment. Smoothing approaches have been investigated as a viable alternative to extended Kalman filter (EKF)-based solutions to the problem. In particular, approaches have been looked at that factorize either the associated information matrix or the measurement Jacobian into square root form. Such techniques have several significant advantages over the EKF: they are faster yet exact; they can be used in either batch or incremental mode; are better equipped to deal with non-linear process and measurement models; and yield the entire robot trajectory, at lower cost for a large class of SLAM problems. In addition, in an indirect but dramatic way, column ordering heuristics automatically exploit the locality inherent in the geographic nature of the SLAM problem. This paper presents the theory underlying these methods, along with an interpretation of factorization in terms of the graphical model associated with the SLAM problem. Both simulation results and actual SLAM experiments in large-scale environments are presented that underscore the potential of these methods as an alternative to EKF-based approaches.

An A Prolog decision support system for the Space Shuttle The goal of this paper is to test if a programming methodology based on the declarative language A-Prolog and the systems for computing answer sets of such programs, can be successfully applied to the development of medium size knowledge-intensive applications. We report on a successful design and development of such a system controlling some of the functions of the Space Shuttle.

Domain adaptation for object recognition: An unsupervised approach Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.

An Unsupervised Feature Learning Approach to Improve Automatic Incident Detection. Sophisticated automatic incident detection (AID) technology plays a key role in contemporary transportation systems. Though many papers were devoted to study incident classification algorithms, few study investigated how to enhance feature representation of incidents to improve AID performance. In this paper, we propose to use an unsupervised feature learning algorithm to generate higher level features to represent incidents. We used real incident data in the experiments and found that effective feature mapping function can be learnt from the data crosses the test sites. With the enhanced features, detection rate (DR), false alarm rate (FAR) and mean time to detect (MTTD) are significantly improved in all of the three representative cases. This approach also provides an alternative way to reduce the amount of labeled data, which is expensive to obtain, required in training better incident classifiers since the feature learning is unsupervised. © 2012 IEEE.

Protecting RAID Arrays against Unexpectedly High Disk Failure Rates Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n^2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.