abstract
stringlengths 383
2.62k
| TLDR
stringlengths 4
250
|
|---|---|
Human language is grounded on multimodal knowledge including visual knowledge like colors, sizes, and shapes. However, current large-scale pre-trained language models rely on the text-only self-supervised training with massive text data, which precludes them from utilizing relevant visual information when necessary. To address this, we propose a novel pre-training framework, named VaLM, to Visually-augment text tokens with retrieved relevant images for Language Modeling. Specifically, VaLM builds on a novel latent text-image alignment method via an image retrieval module to fetch corresponding images given a textual context. With the visually-augmented context, VaLM uses a visual knowledge fusion layer to enable multimodal grounded language modeling by attending on both text context and visual knowledge in images. We evaluate VaLM on various visual knowledge intensive commonsense reasoning tasks, which require visual information to excel. The experimental results illustrate that VaLM outperforms all strong language-only and vision-language baselines with substantial gains on reasoning object commonsense including color, size, and shape. | We propose a novel pre-trained framework, to Visually-augment text tokens with retrieved relevant images for multimodal grounded Language Modeling. |
To understand better good generalization performance in state-of-the-art neural network (NN) models, and in particular the success of the AlphaHat metric based on Heavy-Tailed Self-Regularization (HT-SR) theory, we analyze of a corpus of models that was made publicly-available for a contest to predict the generalization accuracy of NNs. These models include a wide range of qualities and were trained with a range of architectures and regularization hyperparameters. We break AlphaHat into its two subcomponent metrics: a scale-based metric; and a shape-based metric. We identify what amounts to a Simpson’s paradox: where “scale” metrics (from traditional statistical learning theory) perform well in aggregate, but can perform poorly on subpartitions of the data of a given depth, when regularization hyperparameters are varied; and where “shape” metrics (from HT-SR theory) perform well on each subpartition of the data, when hyperparameters are varied for models of a given depth, but can perform poorly overall when models with varying depths are aggregated. Our results highlight the subtlety of comparing models when both architectures and hyperparameters are varied; the complementary role of implicit scale versus implicit shape parameters in understanding NN model quality; and the need to go beyond one-size-fits-all metrics based on upper bounds from generalization theory to describe the performance of NN models. Our results also clarify further why the AlphaHat metric from HT-SR theory works so well at predicting generalization across a broad range of CV and NLP models. | diagnostics for neural network models to understand better their performance |
The ability to predict future visual observations conditioned on past observations and motor commands can enable embodied agents to plan solutions to a variety of tasks in complex environments. This work shows that we can create good video prediction models by pre-training transformers via masked visual modeling. Our approach, named MaskViT, is based on two simple design decisions. First, for memory and training efficiency, we use two types of window attention: spatial and spatiotemporal. Second, during training, we mask a variable percentage of tokens instead of a fixed mask ratio. For inference, MaskViT generates all tokens via iterative refinement where we incrementally decrease the masking ratio following a mask scheduling function. On several datasets we demonstrate that MaskViT outperforms prior works in video prediction, is parameter efficient, and can generate high resolution videos ($256 \times $256). Further, we demonstrate the benefits of inference speedup (up to $512 \times$) due to iterative decoding by using MaskViT for planning on a real robot. Our work suggests that we can endow embodied agents with powerful predictive models by leveraging the general framework of masked visual modeling with minimal domain knowledge. | We propose to learn a Transformer based video prediction model via masked visual modeling. |
Error correction code (ECC) is an integral part of the physical communication layer, ensuring reliable data transfer over noisy channels.
Recently, neural decoders have demonstrated their advantage over classical decoding techniques.
However, recent state-of-the-art neural decoders suffer from high complexity and lack the important iterative scheme characteristic of many legacy decoders.
In this work, we propose to employ denoising diffusion models for the soft decoding of linear codes at arbitrary block lengths.
Our framework models the forward channel corruption as a series of diffusion steps that can be reversed iteratively.
Three contributions are made: (i) a diffusion process suitable for the decoding setting is introduced, (ii) the neural diffusion decoder is conditioned on the number of parity errors, which indicates the level of corruption at a given step, (iii) a line search procedure based on the code's syndrome obtains the optimal reverse diffusion step size.
The proposed approach demonstrates the power of diffusion models for ECC and is able to achieve state-of-the-art accuracy, outperforming the other neural decoders by sizable margins, even for a single reverse diffusion step. | We propose a novel SOTA Neural error correction decoder based on a new diffusion model. |
Training a multi-agent reinforcement learning (MARL) model with sparse reward is notoriously difficult because the terminal reward is induced by numerous interactions among agents. In this study, we propose linear programming-based hierarchical MARL (LPMARL) to learn effective coperative strategy among agents. LPMARL is composed of two hierarchical decision-making schemes: (1) solving an agent-task assignment problem and (2) solving a local cooperative game among agents that are assigned to the same task. For the first step, LPMARL formulates the agent-task assignment problem as linear programming (LP) using the state-dependent cost parameters generated by a graph neural network (GNN). Solving the LP can be considered as assigning tasks to agents, which decomposes the original problem into a set of task-dependent sub-problems. After solving the formulated LP, LPMARL employs a general MARL strategy to derive a lower-level policy to solve each sub-task in a cooperative manner. We train the LP-parameter generating GNN layer and the low-level MARL policy network, which are the essential components for making hierarchical decisions, in an end-to-end manner using the implicit function theorem. We empirically demonstrate that LPMARL learns an optimal agent-task allocation and the subsequent local cooperative control policy among agents in sub-groups for solving various mixed cooperative-competitive environments. | Linear programming-based optimal agent-task allocation for hierarchical multi-agent reinforcement learning. |
Extracting decentralised policies from joint action-values is an attractive way to exploit centralised learning. It is possible to apply monotonic value factorisation to guarantee consistency between the centralised and decentralised policies. However, the best strategy for training decentralised policies when the target joint action-values are non-monotonic and stochastic is still unclear. We propose a novel value factorisation method named uncertainty-based target shaping (UTS) to solve this problem. UTS employs networks that estimate the reward and the following state's embedding, where the large prediction error indicates that the target is stochastic. By replacing deterministic targets for the suboptimal with the best per-agent values, we enforce that all shaped targets become a subset of the space that can be represented by monotonic value factorisation. Empirical results show that UTS outperforms state-of-the-art baselines on multiple benchmarks, including matrix games, predator-prey, and challenging tasks in the StarCraft II micromanagement. | We propose a novel value factorisation method to deal with non-monotonic and stochastic target joint action-values. |
Federated learning is a distributed paradigm that allows multiple parties to collaboratively train deep models without exchanging the raw data. However, the data distribution among clients is naturally non-i.i.d., which leads to severe degradation of the learnt model. The primary goal of this paper is to develop a robust federated learning algorithm to address feature shift in clients’ samples, which can be caused by various factors, e.g., acquisition differences in medical imaging. To reach this goal, we propose FedFA to tackle federated learning from a dis- tinct perspective of federated feature augmentation. FedFA is based on a major insight that each client’s data distribution can be characterized by statistics (i.e., mean and standard deviation) of latent features; and it is likely to manipulate these local statistics globally, i.e., based on information in the entire federation, to let clients have a better sense of the underlying distribution and therefore alleviate local data bias. Based on this insight, we propose to augment each local feature statistic probabilistically based on a normal distribution, whose mean is the original statistic and variance quantifies the augmentation scope. Key to our approach is the determination of a meaningful Gaussian variance, which is accomplished by taking into account not only biased data of each individual client, but also underlying feature statistics characterized by all participating clients. We offer both theoretical and empirical justifications to verify the effectiveness of FedFA. Our code is available at https://github.com/tfzhou/FedFA.
| We propose a simple, flexible, effective, and robust method, named as FedFA, to solve federated learning from a novel perspective of federated feature augmentation. |
In this work, we develop Di-SSL, a Diversity-inducing Self-Supervised Learning technique for histopathology image analysis. SSL techniques, such as contrastive and non-contrastive approaches, have been shown to learn rich and effective rep- resentations without any human supervision. Lately, computational pathology has also benefited from the resounding success of SSL. In this work, we develop a novel domain-aware pretext task to enhance representation learning in digital pathology. Our analysis of vanilla SSL-pretrained models’ attention distribution reveals an insightful observation: sparsity in attention, i.e, models tends to localize most of their attention to some prominent patterns in the image. Although atten- tion sparsity can be beneficial in natural images due to these prominent patterns being the object of interest itself, this can be sub-optimal in digital pathology; this is because, unlike natural images, digital pathology scans are not object-centric, but rather a complex phenotype of various spatially intermixed biological com- ponents. Inadequate diversification of attention in these complex images could result in crucial information loss. To address this, we first leverage cell segmenta- tion to densely extract multiple histopathology-specific representations. We then propose a dense pretext task for SSL, designed to match the multiple correspond- ing representations between the views. Through this, the model learns to attend to various components more closely and evenly, thus inducing adequate diversi- fication in attention for capturing context rich representations. Through quantita- tive and qualitative analysis on multiple slide-level tasks across cancer types, and patch-level classification tasks, we demonstrate the efficacy of our method and observe that the attention is more globally distributed. Specifically, we obtain a relative improvement in accuracy of up to 6.9% in slide-level and 2% in patch level classification tasks (corresponding AUC improvement up to 7.9% and 0.7%, respectively) over a baseline SSL model. | We introduce Di-SSL, a diversity-inducing self-supervised learning method to enhance the representation learning in Digital Pathology. |
Meta-reinforcement learning has widely been used as a learning-to-learn framework to solve unseen tasks with limited experience. However, the aspect of constraint violations has not been adequately addressed in the existing works, making their application restricted in real-world settings. In this paper, we study the problem of meta-safe reinforcement learning (meta-SRL) through the CMDP-within-online framework. We obtain task-averaged regret guarantees for the reward maximization (optimality gap) and constraint violations using gradient-based meta-learning and show that the task-averaged optimality gap and constraint satisfaction improve with task-similarity in the static environment, or task-relatedness in the changing environment. Several technical challenges arise when making this framework practical while still having strong theoretical guarantees. To address these challenges, we propose a meta-algorithm that performs inexact online learning on the upper bounds of intra-task optimality gap and constraint violations estimated by off-policy stationary distribution corrections. Furthermore, we enable the learning rates to be adapted for every task and extend our approach to settings with the dynamically changing task environments. Finally, experiments are conducted to demonstrate the effectiveness of our approach. The proposed theoretical framework is the first to handle the nonconvexity and stochastic nature of within-task CMDPs, while exploiting inter-task dependency for multi-task safe learning.
| We study the problem of meta-reinforcement learning (meta-RL) for constrained Markov decision processes (CMDPs) through the inexact CMDP-within-online framework. |
Estimating the geographical range of a species from in situ observational data is a challenging and important geospatial prediction problem. Given a set of locations indicating where a species has been observed, the goal is to learn a model that can predict how likely it is for the species to be present at any other location. While this is a well-studied problem, traditional approaches are unable to take advantage of more recently available large-scale datasets that cover many locations and species. We propose a new approach that jointly estimates the geographical ranges of tens of thousands of different species simultaneously. We develop a series of benchmark evaluation tasks that measure different aspects of the species range and spatial representation learning problems. We show that our approach scales both in terms of amount of training data and species, where adding more data enables the models to learn better spatial representations that generalize to other species. Despite being only trained on weakly supervised crowdsourced data, our models can approach the predictions of current expert-developed gold standard models. | A new model for jointly estimating the spatial range of thousands of different species from sparse partially observed data. |
Recent large-scale neural autoregressive sequence models have shown impressive performances on a variety of natural language generation tasks. However, their generated sequences often exhibit degenerate properties such as non-termination, undesirable repetition, and premature termination, when generated with decoding algorithms such as greedy search, beam search, top-$k$ sampling, and nucleus sampling. In this paper, we focus on the problem of non-terminating sequences resulting from an incomplete decoding algorithm. We first define an incomplete probable decoding algorithm which includes greedy search, top-$k$ sampling, and nucleus sampling, beyond the incomplete decoding algorithm originally put forward by Welleck et al. (2020). We then propose a non-monotonic self-terminating language model, which significantly relaxes the constraint of monotonically increasing termination probability in the originally proposed self-terminating language model by Welleck et al. (2020), to address the issue of non-terminating sequences when using incomplete probable decoding algorithms. We prove that our proposed model prevents non-terminating sequences when using not only incomplete probable decoding algorithms but also beam search. We empirically validate our model on sequence completion tasks with various architectures. | We propose a new method to prevent language models from non-terminating sequences resulting from incomplete decoding algorithms. |
Machine learning models exhibit two seemingly contradictory phenomena: training data memorization and various forms of forgetting. In memorization, models overfit specific training examples and become susceptible to privacy attacks. In forgetting, examples which appeared early in training are forgotten by the end. In this work, we connect these phenomena.
We propose a technique to measure to what extent models ``forget'' the specifics of training examples, becoming less susceptible to privacy attacks on examples they have not seen recently.
We show that, while non-convexity can prevent forgetting from happening in the worst-case, standard image,speech, and language models empirically do forget examples over time.
We identify nondeterminism as a potential explanation, showing that deterministically trained models do not forget.
Our results suggest that examples seen early when training with extremely large datasets---for instance those examples used to pre-train a model---may observe privacy benefits at the expense of examples seen later. | When models are trained on large datasets, we show that privacy attacks become less effective on examples seen early in training, and investigate why. |
Region proposal networks (RPN) are a key component of modern object detectors. An RPN identifies image boxes likely to contain objects, and so worth further investigation. An RPN false negative is unrecoverable, so the performance of an object detector can be significantly affected by RPN behavior, particularly in low-data regimes. The RPN for a few shot detector is trained on base classes. Our experiments demonstrate that, if the distribution of box aspect ratios for base classes is different from that for novel classes, errors caused by RPN failure to propose a good box become significant. This is predictable: for example, an RPN trained on base classes that are mostly square will tend to miss short wide boxes. It has not been noticed to date because the (relatively few) standard base/novel class splits on current datasets do not display this effect. But changing the base/novel split highlights the problem. We describe datasets where the distribution shift is severe using PASCAL VOC, COCO, and LVIS datasets.
We show that the effect can be mitigated by training multiple distinct but cooperating specialized RPNs. Each specializes in a different aspect ratio, but cooperation constraints reduce the extent to which the RPNs are tuned. This means that if a box is missed by one RPN, it has a good chance of being picked up by another. Experimental evaluation confirms this approach results in substantial improvements in performance on the ARShift benchmarks, while remaining comparable to SOTA on conventional splits. Our approach applies to any few-shot detector and consistently improves performance of detectors. | We propose Cooperating RPN’s for improving the fairness to object aspect ratio distribution in few-shot object detection. |
Many recent works on understanding deep learning try to quantify how much individual data instances influence the optimization and generalization of a model. Such attempts reveal characteristics and importance of individual instances, which may provide useful information in diagnosing and improving deep learning. However, most of the existing works on data valuation require actual training of a model, which often demands high-computational cost. In this paper, we provide a training-free data valuation score, called complexity-gap score, which is a data-centric score to quantify the influence of individual instances in generalization of two-layer overparameterized neural networks. The proposed score can quantify irregularity of the instances and measure how much each data instance contributes in the total movement of the network parameters during training. We theoretically analyze and empirically demonstrate the effectiveness of the complexity-gap score in finding `irregular or mislabeled' data instances, and also provide applications of the score in analyzing datasets and diagnosing training dynamics. Our code is publicly available at https://github.com/JJchy/CG_score. | In this paper, we define a training-free data valuation score, which can be directly computed from data and can effectively quantify the impact of individual instances in optimization and generalization of neural networks. |
Bi-level optimization plays a key role in a lot of machine learning applications. Existing state-of-the-art bi-level optimization methods are limited to smooth or some specific non-smooth lower-level problems. Therefore, achieving an efficient algorithm for the bi-level problems with a generalized non-smooth lower-level objective is still an open problem. To address this problem, in this paper, we propose a new bi-level optimization algorithm based on smoothing and penalty techniques. Using the theory of generalized directional derivative, we derive new conditions for the bilevel optimization problem with nonsmooth, perhaps non-Lipschitz lower-level problem, and prove our method can converge to the points satisfying these conditions. We also compare our method with existing state-of-the-art bi-level optimization methods and demonstrate that our method is superior to the others in terms of accuracy and efficiency. | We propose a method solving the nonsmooth bilevel problem and give a new analysis |
Most of the existing offline reinforcement learning (RL) studies assume the available dataset is sampled directly from the target environment. However, in some practical applications, the available data are often coming from several related but heterogeneous environments. A theoretical understanding of efficient learning from heterogeneous offline datasets remains lacking. In this work, we study the problem of learning a (hidden) underlying Markov decision process (MDP) based on heterogeneous offline datasets collected from multiple randomly perturbed data sources. A novel HetPEVI algorithm is proposed, which jointly considers two types of uncertainties: sample uncertainties from the finite number of data samples per data source, and source uncertainties due to a finite number of data sources. Building on HetPEVI, we further incorporate reference-advantage decompositions and Bernstein-type penalties to propose the HetPEVI-Adv algorithm. Theoretical analysis not only proves the effectiveness of both HetPEVI and HetPEVI-Adv but also demonstrates the advantage of the latter. More importantly, the results explicitly characterize the learning loss due to the finite heterogeneously realized environments compared with sampling directly from the underlying MDP. Finally, we extend the study to MDPs with linear function approximation and propose the HetPEVI-Lin algorithm that provides additional efficiency guarantees beyond the tabular case. | This work investigated the problem of learning an underlying MDP with offline datasets from heterogeneous sources and proposed several provably efficient designs. |
In this paper, we introduce the notion of replicable policies in the context of stochastic bandits, one of the canonical problems in interactive learning. A policy in the bandit environment is called replicable if it pulls, with high probability, the exact same sequence of arms in two different and independent executions (i.e., under independent reward realizations). We show that not only do replicable policies exist, but also they achieve almost the same optimal (non-replicable) regret bounds in terms of the time horizon. More specifically, in the stochastic multi-armed bandits setting, we develop a policy with an optimal problem-dependent regret bound whose dependence on the replicability parameter is also optimal. Similarly, for stochastic linear bandits (with finitely and infinitely many arms) we develop replicable policies that achieve the best-known problem-independent regret bounds with an optimal dependency on the replicability parameter. Our results show that even though randomization is crucial for the exploration-exploitation trade-off, an optimal balance can still be achieved while pulling the exact same arms in two different rounds of executions. | We provide a definition of reproducibility in the context of stochastic bandit problems and we develop algorithms with low regret in various environments. |
Fair representation learning (FRL) is a popular class of methods aiming to produce fair classifiers via data preprocessing. However, recent work has shown that prior methods achieve worse accuracy-fairness tradeoffs than originally suggested by their results. This dictates the need for FRL methods that provide provable upper bounds on unfairness of any downstream classifier, a challenge yet unsolved. In this work we address this challenge and propose Fairness with Restricted Encoders (FARE), the first FRL method with provable fairness guarantees. Our key insight is that restricting the representation space of the encoder enables us to derive suitable fairness guarantees, while allowing empirical accuracy-fairness tradeoffs comparable to prior work. FARE instantiates this idea with a tree-based encoder, a choice motivated by inherent advantages of decision trees when applied in our setting. Crucially, we develop and apply a practical statistical procedure that computes a high-confidence upper bound on the unfairness of any downstream classifier. In our experimental evaluation on several datasets and settings we demonstrate that FARE produces tight upper bounds, often comparable with empirical results of prior methods, which establishes the practical value of our approach.
| We present the first provable fair representation learning method. |
Common training strategies for deep neural networks are computationally expensive, continuing to redundantly train and evaluate on classes already well-understood by the model. A common strategy to diminish this cost is to reduce data used in training, however this often comes at the expense of the model's accuracy or an additional computational cost in training. We propose progressive data dropout (PDD), an adaptive training strategy which performs class-level data dropout from the training set as the network develops an understanding for each class. Our experiments on large-scale image classification demonstrate PDD reduces the total number of datapoints needed to train the network by a factor of 10, reducing the overall training time without significantly impacting accuracy or modifying the model architecture. We additionally demonstrate improvements via experiments and ablations on computer vision benchmarks, including MNIST, Fashion-MNIST, SVHN, CIFAR, and ImageNet datasets. | PDD is a model-agnostic strategy for removing class-level data as learned by the model during training. |
We propose $\textit{EventFormer}$, a computationally efficient event-based representation learning framework for asynchronously processing event camera data. EventFormer treats sparse input events as a spatially unordered set and models their spatial interactions using self-attention mechanism. An associative memory-augmented recurrent module is used to correlate with the stored representation computed from past events. A memory addressing mechanism is proposed to store and retrieve the latent states only $\textit{where}$ these events occur and update them only $\textit{when}$ they occur. The representation learning shift from input space to the latent memory space resulting in reduced computation cost for processing each event. We show that EventFormer achieves 0.5$\%$ and 9$\%$ better accuracy with 30000$\times$ and 200$\times$ less computation compared to the state-of-the-art dense and event-based method, respectively, on event-based object recognition datasets. | We propose EventFormer, an asynchronous spatiotemporal representation learning framework augmented by an associative memory to efficiently perform event-based perception. |
Link prediction is one important application of graph neural networks (GNNs). Most existing GNNs for link prediction are based on one-dimensional Weisfeiler-Lehman ($1$-WL) test. $1$-WL-GNNs first compute node representations by iteratively passing neighboring node features to the center, and then obtain link representations by aggregating the pairwise node representations. As pointed out by previous works, this two-step procedure results in low discriminating power, as $1$-WL-GNNs by nature learn node-level representations instead of link-level. In this paper, we study a completely different approach which can directly obtain node pair (link) representations based on \textit{two-dimensional Weisfeiler-Lehman ($2$-WL) tests}. $2$-WL tests directly use links (2-tuples) as message passing units instead of nodes, and thus can directly obtain link representations. We theoretically analyze the expressive power of $2$-WL tests to discriminate non-isomorphic links, and prove their superior link discriminating power than $1$-WL. Based on different $2$-WL variants, we propose a series of novel $2$-WL-GNN models for link prediction. Experiments on a wide range of real-world datasets demonstrate their competitive performance to state-of-the-art baselines and superiority over plain $1$-WL-GNNs. | We propose provably powerful 2-WL variants for link prediction and successfully implement them to get competitive results and speed advantage. |
In this paper, we propose CLIP-Dissect, a new technique to automatically describe the function of individual hidden neurons inside vision networks. CLIP-Dissect leverages recent advances in multimodal vision/language models to label internal neurons with open-ended concepts without the need for any labeled data or human examples. We show that CLIP-Dissect provides more accurate descriptions than existing methods for last layer neurons where the ground-truth is available as well as qualitatively good descriptions for hidden layer neurons. In addition, our method is very flexible: it is model agnostic, can easily handle new concepts and can be extended to take advantage of better multimodal models in the future. Finally CLIP-Dissect is computationally efficient and can label all neurons from five layers of ResNet-50 in just 4 minutes, which is more than 10$\times$ faster than existing methods. Our code is available at https://github.com/Trustworthy-ML-Lab/CLIP-dissect. | We propose an automated method for generating descriptions of the representation learned by hidden layer neurons, leveraging the multimodal CLIP-model. |
Deep learning has had tremendous success at learning low-dimensional representations of high-dimensional data. This success would be impossible if there was no hidden low-dimensional structure in data of interest; this existence is posited by the manifold hypothesis, which states that the data lies on an unknown manifold of low intrinsic dimension. In this paper, we argue that this hypothesis does not properly capture the low-dimensional structure typically present in image data. Assuming that data lies on a single manifold implies intrinsic dimension is identical across the entire data space, and does not allow for subregions of this space to have a different number of factors of variation. To address this deficiency, we consider the union of manifolds hypothesis, which states that data lies on a disjoint union of manifolds of varying intrinsic dimensions. We empirically verify this hypothesis on commonly-used image datasets, finding that indeed, observed data lies on a disconnected set and that intrinsic dimension is not constant. We also provide insights into the implications of the union of manifolds hypothesis in deep learning, both supervised and unsupervised, showing that designing models with an inductive bias for this structure improves performance across classification and generative modelling tasks. Our code is available at https://github.com/layer6ai-labs/UoMH. | We show data of interest has varying intrinsic dimension, thus conforming to a union of manifolds hypothesis rather than the manifold hypothesis; and we study some implications in deep learning. |
Most existing Image Restoration (IR) models are task-specific, which can not be generalized to different degradation operators. In this work, we propose the Denoising Diffusion Null-Space Model (DDNM), a novel zero-shot framework for arbitrary linear IR problems, including but not limited to image super-resolution, colorization, inpainting, compressed sensing, and deblurring. DDNM only needs a pre-trained off-the-shelf diffusion model as the generative prior, without any extra training or network modifications. By refining only the null-space contents during the reverse diffusion process, we can yield diverse results satisfying both data consistency and realness. We further propose an enhanced and robust version, dubbed DDNM+, to support noisy restoration and improve restoration quality for hard tasks. Our experiments on several IR tasks reveal that DDNM outperforms other state-of-the-art zero-shot IR methods. We also demonstrate that DDNM+ can solve complex real-world applications, e.g., old photo restoration. | We present a novel zero-shot image restoration framework, achieving state-of-the-art performance. |
Distillation-aware Neural Architecture Search (DaNAS) aims to search for an optimal student architecture that obtains the best performance and/or efficiency when distilling the knowledge from a given teacher model. Previous DaNAS methods have mostly tackled the search for the neural architecture for fixed datasets and the teacher, which are not generalized well on a new task consisting of an unseen dataset and an unseen teacher, thus need to perform a costly search for any new combination of the datasets and the teachers. For standard NAS tasks without KD, meta-learning-based computationally efficient NAS methods have been proposed, which learn the generalized search process over multiple tasks (datasets) and transfer the knowledge obtained over those tasks to a new task. However, since they assume learning from scratch without KD from a teacher, they might not be ideal for DaNAS scenarios. To eliminate the excessive computational cost of DaNAS methods and the sub-optimality of rapid NAS methods, we propose a distillation-aware meta-accuracy prediction model, DaSS (Distillation-aware Student Search), which can predict a given architecture's final performances on a dataset when performing KD with a given teacher, without having actually to train it on the target task. The experimental results demonstrate that our proposed meta-prediction model successfully generalizes to multiple unseen datasets for DaNAS tasks, largely outperforming existing meta-NAS methods and rapid NAS baselines. Code is available at https://github.com/CownowAn/DaSS. | We propose a one-shot meta accuracy prediction model which can predict a given architecture's final performances on a dataset when performing KD with a given teacher, without having to actually train it on the target task. |
Knowledge graphs are inherently incomplete. Therefore substantial research has been directed toward knowledge graph completion (KGC), i.e., predicting missing triples from the information represented in the knowledge graph (KG). KG embedding models (KGEs) have yielded promising results for KGC, yet any current KGE is incapable of: (1) fully capturing vital inference patterns (e.g., composition), (2) capturing prominent patterns jointly (e.g., hierarchy and composition), and (3) providing an intuitive interpretation of captured patterns. In this work, we propose ExpressivE, a fully expressive spatio-functional KGE that solves all these challenges simultaneously. ExpressivE embeds pairs of entities as points and relations as hyper-parallelograms in the virtual triple space $\mathbb{R}^{2d}$. This model design allows ExpressivE not only to capture a rich set of inference patterns jointly but additionally to display any supported inference pattern through the spatial relation of hyper-parallelograms, offering an intuitive and consistent geometric interpretation of ExpressivE embeddings and their captured patterns. Experimental results on standard KGC benchmarks reveal that ExpressivE is competitive with state-of-the-art KGEs and even significantly outperforms them on WN18RR. | ExpressivE: A fully expressive KGC model that captures a rich set of patterns with an intuitive geometric interpretation and state-of-the-art performance. |
We study multi-agent general-sum Markov games with nonlinear function approximation. We focus on low-rank Markov games whose transition matrix admits a hidden low-rank structure on top of an unknown non-linear representation. The goal is to design an algorithm that (1) finds an $\varepsilon$-equilibrium policy sample efficiently without prior knowledge of the environment or the representation, and (2) permits a deep-learning friendly implementation. We leverage representation learning and present a model-based and a model-free approach to construct an effective representation from collected data. For both approaches, the algorithm achieves a sample complexity of poly$(H,d,A,1/\varepsilon)$, where $H$ is the game horizon, $d$ is the dimension of the feature vector, $A$ is the size of the joint action space and $\varepsilon$ is the optimality gap. When the number of players is large, the above sample complexity can scale exponentially with the number of players in the worst case. To address this challenge, we consider Markov Games with a factorized transition structure and present an algorithm that escapes such exponential scaling. To our best knowledge, this is the first sample-efficient algorithm for multi-agent general-sum Markov games that incorporates (non-linear) function approximation. We accompany our theoretical result with a neural network-based implementation of our algorithm and evaluate it against the widely used deep RL baseline, DQN with fictitious play. | We provide a general representation learning framework for multi-player general-sum Markov games. |
There is growing interest in understanding how real brains may approximate gradients and how gradients can be used to train neuromorphic chips. However, neither real brains nor neuromorphic chips can perfectly follow the loss gradient, so parameter updates would necessarily use gradient estimators that have some variance and/or bias. Therefore, there is a need to understand better how variance and bias in gradient estimators impact learning dependent on network and task properties. Here, we show that variance and bias can impair learning on the training data, but some degree of variance and bias in a gradient estimator can be beneficial for generalization. We find that the ideal amount of variance and bias in a gradient estimator are dependent on several properties of the network and task: the size and activity sparsity of the network, the norm of the gradient, and the curvature of the loss landscape. As such, whether considering biologically-plausible learning algorithms or algorithms for training neuromorphic chips, researchers can analyze these properties to determine whether their approximation to gradient descent will be effective for learning given their network and task properties. | We characterize the impact of variance and bias in gradient estimates on learning and generalization and study how network architecture properties modulate these effects. |
While recent camera-only 3D detection methods leverage multiple timesteps, the limited history they use significantly hampers the extent to which temporal fusion can improve object perception. Observing that existing works' fusion of multi-frame images are instances of temporal stereo matching, we find that performance is hindered by the interplay between 1) the low granularity of matching resolution and 2) the sub-optimal multi-view setup produced by limited history usage. Our theoretical and empirical analysis demonstrates that the optimal temporal difference between views varies significantly for different pixels and depths, making it necessary to fuse many timesteps over long-term history. Building on our investigation, we propose to generate a cost volume from a long history of image observations, compensating for the coarse but efficient matching resolution with a more optimal multi-view matching setup. Further, we augment the per-frame monocular depth predictions used for long-term, coarse matching with short-term, fine-grained matching and find that long and short term temporal fusion are highly complementary. While maintaining high efficiency, our framework sets new state-of-the-art on nuScenes, achieving first place on the test set and outperforming previous best art by 5.2% mAP and 3.7% NDS on the validation set. Code will be released here: https://github.com/Divadi/SOLOFusion. | We leverage complementary coarse, long-term and fine-grained, short-term multi-view stereo for camera-only 3D object detection. |
Neural Motion Planners (NMPs) have emerged as a promising tool for solving robot navigation tasks in complex environments. However, these methods often require expert data for learning, which limits their application to scenarios where data generation is time-consuming. Recent developments have also led to physics-informed deep neural models capable of representing complex dynamical Partial Differential Equations (PDEs). Inspired by these developments, we propose Neural Time Fields (NTFields) for robot motion planning in cluttered scenarios. Our framework represents a wave propagation model generating continuous arrival time to find path solutions informed by a nonlinear first-order PDE called Eikonal Equation. We evaluate our method in various cluttered 3D environments, including the Gibson dataset, and demonstrate its ability to solve motion planning problems for 4-DOF and 6-DOF robot manipulators where the traditional grid-based Eikonal planners often face the curse of dimensionality. Furthermore, the results show that our method exhibits high success rates and significantly lower computational times than the state-of-the-art methods, including NMPs that require training data from classical planners. | A physics-informed neural time fields model for robot motion planning. |
Reliability of machine learning evaluation --- the consistency of observed evaluation scores across replicated model training runs --- is affected by several sources of nondeterminism which can be regarded as measurement noise. Current tendencies to remove noise in order to enforce reproducibility of research results neglect inherent nondeterminism at the implementation level and disregard crucial interaction effects between algorithmic noise factors and data properties. This limits the scope of conclusions that can be drawn from such experiments. Instead of removing noise, we propose to incorporate several sources of variance, including their interaction with data properties, into an analysis of significance and reliability of machine learning evaluation, with the aim to draw inferences beyond particular instances of trained models. We show how to use linear mixed effects models (LMEMs) to analyze performance evaluation scores, and to conduct statistical inference with a generalized likelihood ratio test (GLRT). This allows us to incorporate arbitrary sources of noise like meta-parameter variations into statistical significance testing, and to assess performance differences conditional on data properties. Furthermore, a variance component analysis (VCA) enables the analysis of the contribution of noise sources to overall variance and the computation of a reliability coefficient by the ratio of substantial to total variance. | Methods for inferential reproducibility of machine learning, using signficance testing under meta-parameter variation, variance components, and reliability coefficients. |
Graph Neural Network (GNN) shows its powerful capability for graph representation learning in various application areas. However, most existing GNN variants learn the graph representations in a non-injective or non-continuous fashion, both reducing the model expressive power. In this paper, we present a theoretical framework to improve the expressive power of GNN by taking both injectivity and continuity into account. Accordingly, we develop \textit{Injective Continuous Graph Neural Network} (ICGNN) that learns the graph and node representations in an injective and continuous fashion, so that it can map similar nodes or graphs to similar embeddings, and non-equivalent nodes or non-isomorphic graphs to different embeddings. We validate the proposed ICGNN model for graph classification and node classification on multiple benchmark datasets including both simple graphs and attributed graphs. The experimental results demonstrate that our model achieves state-of-the-art performances on most of the benchmarks. | graph representation, graph neural network, set representation, expressive power |
Performance predictors can directly predict the performance value of given neural architectures without training, thus broadly being studied to alleviate the prohibitive cost of Neural Architecture Search (NAS). However, existing performance predictors still require training a large number of architectures from scratch to get their performance labels as the training dataset, which is still computationally expensive. To solve this issue, we develop an unsupervised performance predictor called USPP, which can avoid costly dataset construction by using existing fully-trained architectures. Specifically, a progressive domain-invariant feature extraction method is proposed to assist in extracting domain-invariant features due to the great transferability challenge caused by the rich domain-specific features. Furthermore, a learnable representation (denoted as operation embedding) is designed to replace the fixed encoding of the operations to transfer more knowledge about operations to the target search space. In experiments, we train the predictor by the labeled architectures in NAS-Bench-101 and predict the architectures in the DARTS search space. Compared with other state-of-the-art NAS methods, the proposed USPP only costs $0.02$ GPU days but finds the architecture with $97.86\%$ on CIFAR-10 and $96.50\%$ top-1 accuracy on ImageNet. | We propose a performance predictor which can utilize existing fully-trained architectures, thus reducing the high cost of annotating architectures in the background of NAS. |
Many real-world applications of reinforcement learning (RL) require making decisions in continuous action environments. In particular, determining the optimal dose level plays a vital role in developing medical treatment regimes. One challenge in adapting existing RL algorithms to medical applications, however, is that the popular infinite support stochastic policies, e.g., Gaussian policy, may assign riskily high dosages and harm patients seriously. Hence, it is important to induce a policy class whose support only contains near-optimal actions, and shrink the action-searching area for effectiveness and reliability. To achieve this, we develop a novel quasi-optimal learning algorithm, which can be easily optimized in off-policy settings with guaranteed convergence under general function approximations. Theoretically, we analyze the consistency, sample complexity, adaptability, and convergence of the proposed algorithm. We evaluate our algorithm with comprehensive simulated experiments and a dose suggestion real application to Ohio Type 1 diabetes dataset. | The paper proposes a novel learning algorithm for reliable continuous action allocations. |
Post-hoc explanation methods have become increasingly depended upon for understanding black-box classifiers in high-stakes applications, precipitating a need for reliable explanations. While numerous explanation methods have been proposed, recent works have shown that many existing methods can be inconsistent or unstable. In addition, high-performing classifiers are often highly nonlinear and can exhibit complex behavior around the decision boundary, leading to brittle or misleading local explanations. Therefore, there is an impending need to quantify the uncertainty of such explanation methods in order to understand when explanations are trustworthy. We introduce a novel uncertainty quantification method parameterized by a Gaussian Process model, which combines the uncertainty approximation of existing methods with a novel geodesic-based similarity which captures the complexity of the target black-box decision boundary. The proposed framework is highly flexible—it can be used with any black-box classifier and feature attribution method to amortize uncertainty estimates for explanations. We show theoretically that our proposed geodesic-based kernel similarity increases with the complexity of the decision boundary. Empirical results on multiple tabular and image datasets show that our decision boundary-aware uncertainty estimate improves understanding of explanations as compared to existing methods | We introduce a method that generates uncertainty estimates for feature attribution explanations. |
Despite the success of Transformers in various applications from text, vision, and speech domains, they are yet to become standard architectures for mobile and edge device applications due to their heavy memory and computational requirements. While there exist many different approaches to reduce the complexities of the Transformers, such as the pruning of the weights/attentions/tokens, quantization, and distillation, we focus on token pruning, which reduces not only the complexity of the attention operations, but also the linear layers, which have non-negligible computational costs. However, previous token pruning approaches often remove tokens during the feed-forward stage without consideration of their impact on later layers' attentions, which has a potential risk of dropping out important tokens for the given task. To tackle this issue, we propose an attention back-tracking method that tracks the importance of each attention in a Transformer architecture from the outputs to the inputs, to preserve the tokens that have a large impact on the final predictions. We experimentally validate the effectiveness of the method on both NLP and CV benchmarks, using Transformer architectures for both domains, and the results show that the proposed attention back-tracking allows the model to better retain the full models' performance even at high sparsity rates, significantly outperforming all baselines. Qualitative analysis of the examples further shows that our method does preserve semantically meaningful tokens. | We propose an attention back-tracking method that tracks the importance of each attention in a Transformer architecture from the outputs to the inputs, to preserve the tokens that have large impact on the final predictions. |
Large language models (LLMs) have shown increasing in-context learning capabilities through scaling up model and data size. Despite this progress, LLMs are still unable to solve algorithmic reasoning problems. While providing a rationale with the final answer has led to further improvements in multi-step reasoning problems, Anil et al. 2022 showed that even simple algorithmic reasoning tasks such as parity are far from solved. In this work, we identify and study four key stages for successfully teaching algorithmic reasoning to LLMs: (1) formulating algorithms as skills, (2) teaching multiple skills simultaneously (skill accumulation), (3) teaching how to combine skills (skill composition) and (4) teaching how to use skills as tools. We show that it is possible to teach algorithmic reasoning to LLMs via in-context learning, which we refer to as Algorithmic Prompting. We evaluate our approach on a variety of arithmetic and quantitative reasoning tasks, and demonstrate significant boosts in performance over existing prompting techniques. In particular, for long parity, addition, multiplication and subtraction and parity tasks, we achieve an error reduction of approximately 10x, 9x, 5x and 2x respectively compared to the best available baselines. | We study how to teach algorithmic reasoning to LLMs via in-context learning. We show that algorithmic reasoning can be taught by increasing specificity in the way we explain the steps of an algorithm along with running examples. |
The high effectiveness of neural models of code, such as OpenAI Codex and AlphaCode, suggests coding capabilities of models that are at least comparable to those of humans. However, previous work has only used these models for their raw completion, ignoring how the model reasoning, in the form of attention weights, can be used for other downstream tasks. Disregarding the attention weights means discarding a considerable portion of what those models compute when queried. To profit more from the knowledge embedded in these large pre-trained models, this work compares multiple approaches to post-process these valuable attention weights for supporting code exploration. Specifically, we compare to which extent the transformed attention signal of CodeGen, a large and publicly available pre-trained neural model, agrees with how developers look at and explore code when each answering the same sense-making questions about code. At the core of our experimental evaluation, we collect, manually annotate, and open-source a novel eye-tracking dataset comprising 25 developers answering sense-making questions on code over 92 sessions. We empirically evaluate five attention-agnostic heuristics and ten attention-based post processing approaches of the attention signal against our ground truth of developers exploring code, including the novel concept of follow-up attention which exhibits the highest agreement. Beyond the dataset contribution and the empirical study, we also introduce a novel practical application of the attention signal of pre-trained models with completely analytical solutions, going beyond how neural models’ attention mechanisms have traditionally been used.
| We compare how developers and GPT-like language models navigate snippets of source code. |
A long-sought property of machine learning systems is the ability to compose learned concepts in novel ways that would enable them to make sense of new situations. Such capacity for imagination -- a core aspect of human intelligence -- is not yet attained for machines. In this work, we show that object-centric inductive biases can be leveraged to derive an imagination-based learning framework that achieves compositional generalization on a series of tasks. Our method, denoted Object-centric Compositional IMagination (OCIM), decomposes visual reasoning tasks into a series of primitives applied to objects without using a domain-specific language. We show that these primitives can be recomposed to generate new imaginary tasks. By training on such imagined tasks, the model learns to reuse the previously-learned concepts to systematically generalize at test time. We test our model on a series of arithmetic tasks where the model has to infer the sequence of operations (programs) applied to a series of inputs. We find that imagination is key for the model to find the correct solution for unseen combinations of operations. | Our model leveragse object-centric inductive biases to derive an imagination-based learning framework. We show that it leads to better compositional generalization in a visual abstact reasoning task. |
Learning policies via preference-based reward learning is an increasingly popular method for customizing agent behavior, but has been shown anecdotally to be prone to spurious correlations and reward hacking behaviors. While much prior work focuses on causal confusion in reinforcement learning and behavioral cloning, we focus on a systematic study of causal confusion and reward misidentification when learning from preferences. In particular, we perform a series of sensitivity and ablation analyses on several benchmark domains where rewards learned from preferences achieve minimal test error but fail to generalize to out-of-distribution states---resulting in poor policy performance when optimized. We find that the presence of non-causal distractor features, noise in the stated preferences, and partial state observability can all exacerbate reward misidentification. We also identify a set of methods with which to interpret misidentified learned rewards. In general, we observe that optimizing misidentified rewards drives the policy off the reward's training distribution, resulting in high predicted (learned) rewards but low true rewards. These findings illuminate the susceptibility of preference learning to reward misidentification and causal confusion---failure to consider even one of many factors can result in unexpected, undesirable behavior. | We identify and analyze important factors that influence causal confusion when learning rewards from human preference labels. |
Natural data observed in $\mathbb{R}^n$ is often constrained to an $m$-dimensional manifold $\mathcal{M}$, where $m < n$. Current probabilistic models represent this manifold by mapping an $m$-dimensional latent variable through a neural network $f_\theta: \mathbb{R}^m \to \mathbb{R}^n$. Such procedures, which we call pushforward models, incur a straightforward limitation: manifolds cannot in general be represented with a single parameterization, meaning that attempts to do so will incur either computational instability or the inability to learn probability densities within the manifold. To remedy this problem, we propose to model $\mathcal{M}$ as a neural implicit manifold: the set of zeros of a neural network. To learn the data distribution within $\mathcal{M}$, we introduce constrained energy-based models, which use a constrained variant of Langevin dynamics to train and sample within a learned manifold. The resulting model can be manipulated with an arithmetic of manifolds, which allows practitioners to take unions and intersections of model manifolds. In experiments on synthetic and natural data, we show that constrained EBMs can learn manifold-supported distributions with complex topologies more accurately than pushforward models. | We propose a new model for probability distributions on topologically complex data manifolds which learns manifolds implicitly as the set of zeros of a neural network and then learns the distribution within using a constrained energy-based model. |
Large Transformer-based Pretrained Language Models (PLMs) dominate almost all Natural Language Processing (NLP) tasks. Nevertheless, they still make mistakes from time to time. For a model deployed in an industrial environment, fixing these mistakes quickly and robustly is vital to improve user experiences. Previous works formalize such problems as Model Editing (ME) and mostly focus on fixing one mistake. However, the one-mistake-fixing scenario is not an accurate abstraction of the real-world challenge. In the deployment of AI services, there are ever-emerging mistakes, and the same mistake may recur if not corrected in time. Thus a preferable solution is to rectify the mistakes as soon as they appear nonstop. Therefore, we extend the existing ME into the Sequential Model Editing (SME) to help develop more practical editing methods. Our study shows that current ME methods either fail to make a sequence of edits or to remember previous edits. We then introduce Transformer-Patcher, a novel model editor that can shift the behavior of transformer-based models by simply adding and training a few neurons in the last Feed-Forward Network layer. Experimental results on both classification and generation tasks show that Transformer-Patcher can successively correct up to thousands of errors (Reliability) and generalize to their equivalent inputs (Generality) while retaining the model’s accuracy on irrelevant inputs (Locality). Our method outperforms previous fine-tuning and HyperNetwork-based methods and achieves state-of-the-art performance for Sequential Model Editing (SME). | A Sequential Model Editor to correct model's output on the specific input. |
We introduce new differentially private (DP) mechanisms for gradient-based machine learning (ML) training involving multiple passes (epochs) of a dataset, substantially improving the achievable privacy-utility-computation tradeoffs. Our key contribution is an extension of the online matrix factorization DP mechanism to multiple participations, substantially generalizing the approach of DMRST2022. We first give a non-trivial reduction of the problem with per-iteration vector contributions to the simpler one of scalar contributions. Using this, we formulate the construction of optimal (in total squared error at each iterate) matrix mechanisms for SGD variants as a convex program. We provide a closed form solution to the dual function, leading directly to an efficient optimization algorithms.
While tractable, both solving the convex problem offline and computing the necessary noise masks during training can become prohibitively expensive when many training steps are necessary. To address this, we design a Fourier-transform-based mechanism with significantly less computation and only a minor utility decrease.
Extensive empirical evaluation on two tasks, example-level DP for image classification and user-level DP for language modeling, demonstrate substantial improvements over the previous state-of-the-art. Though our primary application is to ML, we note our main DP results are applicable to arbitrary linear queries and hence may have much broader applicability. | We enable generation of factorization-based methods under multi-participations, which enables us to achieve new SOTA results in private training without amplification. |
We present Phenaki, a model capable of realistic video synthesis given a sequence of textual prompts. Generating videos from text is particularly challenging due to the computational cost, limited quantities of high quality text-video data and variable length of videos. To address these issues, we introduce a new causal model for learning video representation which compresses the video to a small discrete tokens representation. This tokenizer is auto-regressive in time, which allows it to work with video representations of different length.
To generate video tokens from text we are using a bidirectional masked transformer conditioned on pre-computed text tokens. The generated video tokens are subsequently de-tokenized to create the actual video. To address data issues, we demonstrate how joint training on a large corpus of image-text pairs as well as a smaller number of video-text examples can result in generalization beyond what is available in the video datasets. Compared to the previous video generation methods, Phenaki can generate arbitrary long videos conditioned on a sequence of prompts (i.e. time variable text or story) in open domain. To the best of our knowledge, this is the first time a paper studies generating videos from time variable prompts. | Generating videos from text, with prompts that can change over time, and videos that can be as long as multiple minutes. |
Offline reinforcement learning (RL), which aims at learning good policies from historical data, has received significant attention over the past years. Much effort has focused on improving offline RL practicality by addressing the prevalent issue of partial data coverage through various forms of conservative policy learning. While the majority of algorithms do not have finite-
sample guarantees, several provable conservative offline RL algorithms are designed and analyzed within the single-policy concentrability framework that handles partial coverage. Yet, in the nonlinear function approximation setting where confidence intervals are difficult to obtain, existing provable algorithms suffer from computational intractability, prohibitively strong assumptions, and suboptimal statistical rates. In this paper, we leverage the marginalized importance sampling (MIS) formulation of RL and present the first set of offline RL algorithms that are statistically optimal and practical under general function approximation and single-policy concentrability, bypassing the need for uncertainty quantification. We identify that the key to successfully solving the sample-based approximation of the MIS problem is ensuring that certain occupancy validity constraints are nearly satisfied. We enforce these constraints by a novel application of the augmented Lagrangian method and prove the following result: with the MIS formulation, augmented Lagrangian is enough for statistically optimal offline RL. In stark contrast to prior algorithms that induce additional conservatism through methods such as behavior regularization, our approach provably eliminates this need and reinterprets regularizers as "enforcers of occupancy validity" than "promoters of conservatism." | We present practical and statistically optimal offline RL algorithms under general function approximation and single-policy concentrability. |
Preference-based reinforcement learning (RL) provides a framework to train agents using human preferences between two behaviors. However, preference-based RL has been challenging to scale since it requires a large amount of human feedback to learn a reward function aligned with human intent. In this paper, we present Preference Transformer, a neural architecture that models human preferences using transformers. Unlike prior approaches assuming human judgment is based on the Markovian rewards which contribute to the decision equally, we introduce a new preference model based on the weighted sum of non-Markovian rewards. We then design the proposed preference model using a transformer architecture that stacks causal and bidirectional self-attention layers. We demonstrate that Preference Transformer can solve a variety of control tasks using real human preferences, while prior approaches fail to work. We also show that Preference Transformer can induce a well-specified reward and attend to critical events in the trajectory by automatically capturing the temporal dependencies in human decision-making. Code is available on the project website: https://sites.google.com/view/preference-transformer. | We introduce a transformer-based architecture for preference-based RL considering non-Markovian rewards. |
Many empirical studies have demonstrated the performance benefits of conditional computation in neural networks, including reduced inference time and power consumption. We study the fundamental limits of neural conditional computation from the perspective of memorization capacity. For Rectified Linear Unit (ReLU) networks without conditional computation, it is known that memorizing a collection of $n$ input-output relationships can be accomplished via a neural network with $O(\sqrt{n})$ neurons. Calculating the output of this neural network can be accomplished using $O(\sqrt{n})$ elementary arithmetic operations of additions, multiplications and comparisons for each input. Using a conditional ReLU network, we show that the same task can be accomplished using only $O(\log n)$ operations per input. This represents an almost exponential improvement as compared to networks without conditional computation. We also show that the $\Theta(\log n)$ rate is the best possible. Our achievability result utilizes a general methodology to synthesize a conditional network out of an unconditional network in a computationally-efficient manner, bridging the gap between unconditional and conditional architectures. | In classical "unconditional" ReLU nets, one needs $O(\sqrt{n})$ arithmetic operations to recall any one of $n$ stored patterns. Conditional computation reduces this to $O(\log n)$ and this is the best possible. |
We present the Additive Poisson Process (APP), a novel framework that can model the higher-order interaction effects of the intensity functions in Poisson processes using projections into lower-dimensional space. Our model combines the techniques in information geometry to model higher-order interactions on a statistical manifold and in generalized additive models to use lower-dimensional projections to overcome the effects from the curse of dimensionality. Our approach solves a convex optimization problem by minimizing the KL divergence from a sample distribution in lower-dimensional projections to the distribution modeled by an intensity function in the Poisson process. Our empirical results show that our model is able to use samples observed in the lower dimensional space to estimate the higher-order intensity function with extremely sparse observations. | An efficient technique that uses a log-linear model on a partial order structure to approximate a high-dimensional intensity functions in a Poisson Process. |
We introduce a novel machine learning optimizer called LODO, which online meta-learns an implicit inverse Hessian of the loss as a subroutine of quasi-Newton optimization. Our optimizer merges Learning to Optimize (L2O) techniques with quasi-Newton methods to learn neural representations of symmetric matrix vector products, which are more flexible than those in other quasi-Newton methods. Unlike other L2O methods, ours does not require any meta-training on a training task distribution, and instead learns to optimize on the fly while optimizing on the test task, adapting to the local characteristics of the loss landscape while traversing it. Theoretically, we show that our optimizer approximates the inverse Hessian in noisy loss landscapes and is capable of representing a wide range of inverse Hessians. We experimentally verify our algorithm's performance in the presence of noise, and show that simpler alternatives for representing the inverse Hessians worsen performance. Lastly, we use our optimizer to train a semi-realistic deep neural network with 95k parameters, and obtain competitive results against standard neural network optimizers. | We introduce a novel machine learning optimizer which combines learning to optimize with quasi-Newton methodology. |
Out-of-distribution (OOD) detection plays a crucial role in ensuring the safe deployment of deep neural network (DNN) classifiers.
While a myriad of methods have focused on improving the performance of OOD detectors, a critical gap remains in interpreting their decisions.
We help bridge this gap by providing explanations for OOD detectors based on learned high-level concepts.
We first propose two new metrics for assessing the effectiveness of a particular set of concepts for explaining OOD detectors: 1) $\textit{detection completeness}$, which quantifies the sufficiency of concepts for explaining an OOD-detector's decisions, and 2) $\textit{concept separability}$, which captures the distributional separation between in-distribution and OOD data in the concept space.
Based on these metrics, we propose a framework for learning a set of concepts that satisfy the desired properties of detection completeness and concept separability, and demonstrate the framework's effectiveness in providing concept-based explanations for diverse OOD detection techniques.
We also show how to identify prominent concepts that contribute to the detection results via a modified Shapley value-based importance score. | We propose the first work to provide concept-based explanations for out-of-distribution detectors. |
Forming a molecular candidate set that contains a wide range of potentially effective compounds is crucial to the success of drug discovery. While most databases and machine-learning-based generation models aim to optimize particular chemical properties, there is limited literature on how to properly measure the coverage of the chemical space by those candidates included or generated. This problem is challenging due to the lack of formal criteria to select good measures of the chemical space. In this paper, we propose a novel evaluation framework for measures of the chemical space based on two analyses: an axiomatic analysis with three intuitive axioms that a good measure should obey, and an empirical analysis on the correlation between a measure and a proxy gold standard. Using this framework, we are able to identify #Circles, a new measure of chemical space coverage, which is superior to existing measures both analytically and empirically. We further evaluate how well the existing databases and generation models cover the chemical space in terms of #Circles. The results suggest that many generation models fail to explore a larger space over existing databases, which leads to new opportunities for improving generation models by encouraging exploration.
| We propose a novel evaluation framework for measures of the chemical space in the context of drug discovery. |
We study the Trojan Attack problem, where malicious attackers sabotage deep
neural network models with poisoned training data. In most existing works, the
effectiveness of the attack is largely overlooked; many attacks can be ineffective
or inefficient for certain training schemes, e.g., adversarial training. In this paper,
we adopt a novel perspective and look into the quantitative relationship between a
clean model and its Trojaned counterpart. We formulate a successful attack using
classic machine learning language. Under mild assumptions, we show theoretically
that there exists a Trojaned model, named Trojaned Twin, that is very close to the
clean model. This attack can be achieved by simply using a universal Trojan trigger
intrinsic to the data distribution. This has powerful implications in practice; the
Trojaned twin model has enhanced attack efficacy and strong resiliency against
detection. Empirically, we show that our method achieves consistent attack efficacy
across different training schemes, including the challenging adversarial training
scheme. Furthermore, this Trojaned twin model is robust against SoTA
detection methods | A mathematical model for backdoor attack, show the existence of a trojaned twin model of a clean model |
Well-known robust aggregation schemes in federated learning (FL) are shown to be vulnerable to an informed adversary who can tailor training-time attacks [Fang et al., Xie et al.]. We frame robust distributed learning problem as a game between a server and an adversary that is able to optimize strong training-time attacks. We introduce RobustTailor, a simulation-based framework that prevents the adversary from being omniscient. The simulated game we propose enjoys theoretical guarantees through a regret analysis. RobustTailor improves robustness to training-time attacks significantly while preserving almost the same privacy guarantees as standard robust aggregation schemes in FL. Empirical results under challenging attacks show that RobustTailor performs similar to an upper bound with perfect knowledge of honest clients. | We frame robust distributed learning problem as a game between a server and an adversary that optimizes strong training-time attacks. |
Graph Neural Networks (GNNs) have demonstrated powerful representation capability in semi-supervised node classification. In this task, there are often three types of information -- graph structure, node features, and node labels. Existing GNNs usually leverage both node features and graph structure by feature transformation and aggregation, following end-to-end training via node labels. In this paper, we change our perspective by considering these three types of information as three views of nodes. This perspective motivates us to design a new GNN framework as multi-view learning which enables alternating optimization training instead of end-to-end training, resulting in significantly improved computation and memory efficiency. Extensive experiments with different settings demonstrate the effectiveness and efficiency of the proposed method.
| A new Multi-View Learning perspective on GNN, which achieves better computation and memory efficiency. |
We propose SPRINT, an approach for scalable offline policy pre-training based on natural language instructions. SPRINT pre-trains an agent’s policy to execute a diverse set of semantically meaningful skills that it can leverage to learn new tasks faster. Prior work on offline pre-training required tedious manual definition of pre-training tasks or learned semantically meaningless skills via random goal-reaching. Instead, our approach SPRINT (Scalable Pre-training via Relabeling Language INsTructions) leverages natural language instruction labels on offline agent experience, collected at scale (e.g., via crowd-sourcing), to define a rich set of tasks with minimal human effort. Furthermore, by using natural language to define tasks, SPRINT can use pre-trained large language models to automatically expand the initial task set. By relabeling and aggregating task instructions, even across multiple training trajectories, we can learn a large set of new skills during pre-training. In experiments using a realistic household simulator, we show that agents pre-trained with SPRINT learn new long-horizon household tasks substantially faster than with previous pre-training approaches. | We propose a scalable offline policy pre-training approach based on natural language instructions that automatically generates new pre-training tasks with language-model relabeling. |
In order to develop trustworthy downstream applications for semantic segmentation models, it is important to not only understand the performance of a model on datasets, but to localize areas where the model may produce errors.
Pixel-wise error prediction of semantic segmentation maps is a challenging problem in which prior work relies on complicated image resynthesis pipelines.
We introduce \it{error augmentation}, a framework which enables us to learn robust error detectors by applying data transformations independently on the predicted segmentation maps.
This approach enables direct prediction of pixel-wise error in semantic segmentation maps, an approach explored as a naive baseline in prior works, to achieve state of the art performance.
As a proof-of-concept we propose a series of three simple transformations that generate challenging segmentation errors by swapping pixel predictions within a segmentation map.
Our approach outperforms previous methods of error detection for semantic segmentation across all metrics and improves performance by over $7.8\%$ on AUPR-Error.
Additionally, we show that our approach not only generalizes to unseen test examples, but remains reliable despite significant shifts in the target domain.
| Introduces Error Augmentation as a framework for reliably producing error detectors for semantic segmentation with less architectural and computational complexity. |
Federated learning (FL) provides an effective collaborative training paradigm, allowing local agents to train a global model jointly without sharing their local data to protect privacy.
On the other hand, due to the heterogeneous nature of local agents, it is challenging to optimize or even define the fairness for agents, which may discourage valuable participation. For instance, the trained global model may sacrifice the performance of a minority user with high-quality data based on loss optimization over all users.
Existing work usually considers accuracy equity as fairness for different users in FL, which is limited especially under the heterogeneous setting, since it is intuitively "unfair" that agents with low-quality data would achieve similar accuracy.
In this work, we aim to address such limitations and propose a formal fairness definition in FL, fairness via agent-awareness (FAA), which takes the heterogeneous data contributions of local agents into account. In addition, we propose a fair FL training algorithm based on agent clustering (FOCUS) to achieve FAA. Theoretically, we prove the convergence and optimality of FOCUS under mild conditions for linear and general convex loss functions with bounded smoothness. We also prove that FOCUS always achieves higher fairness measured by FAA compared with standard FedAvg protocol under both linear and general convex loss functions. Empirically, we evaluate FOCUS on four datasets, including synthetic data, images, and texts under different settings, and we show that FOCUS achieves significantly higher fairness based on FAA while maintaining similar or even higher prediction accuracy compared with FedAvg and other existing fair FL algorithms.
| We propose a formal definition of fairness via agent-awareness for FL (FAA) on heterogeneous data and a fair FL training algorithm based on agent clustering (FOCUS) to achieve FAA. |
Time-series forecasting has caught increasing attention in the AI research field due to its importance in solving real-world problems across different domains, such as energy, weather, traffic, and economy. As shown in various types of data, it has been a must-see issue to deal with drastic changes, temporal patterns, and shapes in sequential data that previous models are weak in prediction. This is because most cases in time-series forecasting aim to minimize $L_p$ norm distances as loss functions, such as mean absolute error (MAE) or mean square error (MSE). These loss functions are vulnerable to not only considering temporal dynamics modeling but also capturing the shape of signals. In addition, these functions often make models misbehave and return uncorrelated results to the original time-series. To become an effective loss function, it has to be invariant to the set of distortions between two time-series data instead of just comparing exact values. In this paper, we propose a novel loss function, called TILDE-Q (Transformation Invariant Loss function with Distance EQuilibrium), that not only considers the distortions in amplitude and phase but also allows models to capture the shape of time-series sequences. In addition, TILDE-Q supports modeling periodic and non-periodic temporal dynamics at the same time. We evaluate the effectiveness of TILDE-Q by conducting extensive experiments with respect to periodic and non-periodic conditions of data, from naive models to state-of-the-art models. The experiment results indicate that the models trained with TILDE-Q outperform those trained with other training metrics (e.g., MSE, dynamic time warping (DTW), temporal distortion index (TDI), and longest common subsequence (LCSS)). | We designed a novel, lightweight, and shape-aware loss function for time-series forecasting. |
Concept-based interpretations of black-box models are often more intuitive for humans to understand. The most widely adopted approach for concept-based, gradient interpretation is Concept Activation Vector (CAV). CAV relies on learning a linear relation between some latent representation of a given model and concepts. The premise of meaningful concepts lying in a linear subspace of model layers is usually implicitly assumed but does not hold true in general. In this work we proposed Concept Gradient (CG), which extends concept-based, gradient interpretation methods to non-linear concept functions. We showed that for a general (potentially non-linear) concept, we can mathematically measure how a small change of concept affects the model’s prediction, which is an extension of gradient-based interpretation to the concept space. We demonstrated empirically that CG outperforms CAV in attributing concept importance on real world datasets and performed case study on a medical dataset. The code is available at github.com/jybai/concept-gradients. | Extending concept-based gradient interpretation to non-linear concept functions. |
Estimating fluid dynamics is classically done through the simulation and integration of numerical models solving the Navier-Stokes equations, which is computationally complex and time-consuming even on high-end hardware. This is a notoriously hard problem to solve, which has recently been addressed with machine learning, in particular graph neural networks (GNN) and variants trained and evaluated on datasets of static objects in static scenes with fixed geometry. We attempt to go beyond existing work in complexity and introduce a new model, method and benchmark. We propose EAGLE: a large-scale dataset of ∼1.1 million 2D meshes resulting from simulations of unsteady fluid dynamics caused by a moving flow source interacting with nonlinear scene structure of varying geometries, with 600 different scenes of three different types in total. To perform future forecasting of pressure and velocity on the challenging EAGLE dataset, we introduce a new mesh transformer. It leverages node clustering, graph pooling and global attention to learn long-range dependencies between spatially distant data points without needing a large number of iterations, as existing GNN methods do. We show that our transformer outperforms state-of-the-art performance on, both, existing synthetic and real datasets and on EAGLE. Finally, we highlight that our approach learns to attend to airflow, integrating complex information in a single iteration. | We introduce a new large-scale dataset for learning non-steady fluid mechanics and a method based on self-attention on graphs |
Online learning in multi-armed bandits has been a rich area of research for decades, resulting in numerous \enquote{no-regret} algorithms that efficiently learn the arm with highest expected reward. However, in many settings the final decision of which arm to pull isn't under the control of the algorithm itself. For example, a driving app typically suggests a subset of routes (arms) to the driver, who ultimately makes the final choice about which to select. Typically, the human also wishes to learn the optimal arm based on historical reward information, but decides which arm to pull based on a potentially different objective function, such as being more or less myopic about exploiting near-term rewards. In this paper, we show when this joint human-algorithm system can achieve good performance. Specifically, we explore multiple possible frameworks for human objectives and give theoretical regret bounds for regret. Finally, we include experimental results exploring how regret varies with the human decision-maker's objective, as well as the number of arms presented. | Analysis of joint human-algorithm multi-armed bandit systems: human picks the final arms from a subset the algorithm selects. |
The ability to ensure that a classifier gives reliable confidence scores is essential to ensure informed decision-making. To this end, recent work has focused on miscalibration, i.e., the over or under confidence of model scores. Yet calibration is not enough: even a perfectly calibrated classifier with the best possible accuracy can have confidence scores that are far from the true posterior probabilities. This is due to the grouping loss, created by samples with the same confidence scores but different true posterior probabilities. Proper scoring rule theory shows that given the calibration loss, the missing piece to characterize individual errors is the grouping loss. While there are many estimators of the calibration loss, none exists for the grouping loss in standard settings. Here, we propose an estimator to approximate the grouping loss. We show that modern neural network architectures in vision and NLP exhibit grouping loss, notably in distribution shifts settings, which highlights the importance of pre-production validation. | We provide an estimator to evaluate confidence scores beyond calibration, revealing the subgroups heterogeneities that undermine individual predicted probabilities. |
Although Transformer networks outperform various natural language processing tasks, many aspects of their theoretical nature are still unclear. On the other hand, fully connected neural networks have been extensively studied in terms of their approximation and estimation capability where the target function is included in such function classes as H\"older class and Besov class. In this paper, we study the approximation and estimation error of Transformer networks in a setting where the target function takes a fixed-length sentence as an input and belongs to two variants of Besov spaces known as anisotropic Besov and mixed smooth Besov spaces, in which it is shown that Transformer networks can avoid curse of dimensionality. By overcoming the difficulties in limited interactions among tokens, we prove that Transformer networks can accomplish minimax optimal rate. Our result also shows that token-wise parameter sharing in Transformer networks decreases dependence of the network width on the input length. Moreover, we prove that, under suitable situations, Transformer networks dynamically select tokens to pay careful attention to. This phenomenon matches attention mechanism, on which Transformer networks are based. Our analyses strongly support the reason why Transformer networks have outperformed various natural language processing tasks from a theoretical perspective. | This paper is written about the approximation ability of Transformers for functions with various smoothness, and, from a point of view of apporoximation, we prove the token extraction property of Transformers. |
Byzantine-robustness has been gaining a lot of attention due to the growth of the interest in collaborative and federated learning. However, many fruitful directions, such as the usage of variance reduction for achieving robustness and communication compression for reducing communication costs, remain weakly explored in the field. This work addresses this gap and proposes Byz-VR-MARINA -- a new Byzantine-tolerant method with variance reduction and compression. A key message of our paper is that variance reduction is key to fighting Byzantine workers more effectively. At the same time, communication compression is a bonus that makes the process more communication efficient. We derive theoretical convergence guarantees for Byz-VR-MARINA outperforming previous state-of-the-art for general non-convex and Polyak-Lojasiewicz loss functions. Unlike the concurrent Byzantine-robust methods with variance reduction and/or compression, our complexity results are tight and do not rely on restrictive assumptions such as boundedness of the gradients or limited compression. Moreover, we provide the first analysis of a Byzantine-tolerant method supporting non-uniform sampling of stochastic gradients. Numerical experiments corroborate our theoretical findings. | We propose a new Byzantine-tolerant method with variance reduction, communication compression, and theoretical guarantees superior to previously known results |
When deployed for risk-sensitive tasks, deep neural networks must be able to detect instances with labels from outside the distribution for which they were trained.
In this paper we present a novel framework to benchmark the ability of image classifiers to detect class-out-of-distribution instances
(i.e., instances whose true labels do not appear in the training distribution) at various levels of detection difficulty.
We apply this technique to ImageNet, and benchmark 525 pretrained, publicly available, ImageNet-1k classifiers.
The code for generating a benchmark for any ImageNet-1k classifier, along with the benchmarks prepared for the above-mentioned 525 models is available at https://github.com/mdabbah/COOD_benchmarking.
The usefulness of the proposed framework and its advantage over alternative existing benchmarks is demonstrated by analyzing the results obtained for these models, which reveals numerous novel observations including:
(1) knowledge distillation consistently improves class-out-of-distribution (C-OOD) detection performance; (2) a subset of ViTs performs better C-OOD detection than any other model; (3) the language–-vision CLIP model achieves good zero-shot detection performance, with its best instance outperforming 96% of all other models evaluated; (4) accuracy and in-distribution ranking are positively correlated to C-OOD detection; and
(5) we compare various confidence functions for C-OOD detection.
Our companion paper, also published in ICLR 2023 (What Can We Learn From The Selective Prediction And Uncertainty Estimation Performance Of 523 Imagenet Classifiers), examines the uncertainty estimation performance (ranking, calibration, and selective prediction performance) of these classifiers in an in-distribution setting. | We present a framework for benchmarking the performance of image classifiers in detecting OOD. We apply it to benchmark 525 pretrained ImageNet classifiers, and analyze their performance resulting in interesting conclusions |
The recent 3D medical ViTs (e.g., SwinUNETR) achieve the state-of-the-art performances on several 3D volumetric data benchmarks, including 3D medical image segmentation. Hierarchical transformers (e.g., Swin Transformers) reintroduced several ConvNet priors and further enhanced the practical viability of adapting volumetric segmentation in 3D medical datasets. The effectiveness of hybrid approaches is largely credited to the large receptive field for non-local self-attention and the large number of model parameters. We hypothesize that volumetric ConvNets can simulate the large receptive field behavior of these learning approaches with fewer model parameters using depth-wise convolution. In this work, we propose a lightweight volumetric ConvNet, termed 3D UX-Net, which adapts the hierarchical transformer using ConvNet modules for robust volumetric segmentation. Specifically, we revisit volumetric depth-wise convolutions with large kernel (LK) size (e.g. starting from $7\times7\times7$) to enable the larger global receptive fields, inspired by Swin Transformer. We further substitute the multi-layer perceptron (MLP) in Swin Transformer blocks with pointwise depth convolutions and enhance model performances with fewer normalization and activation layers, thus reducing the number of model parameters. 3D UX-Net competes favorably with current SOTA transformers (e.g. SwinUNETR) using three challenging public datasets on volumetric brain and abdominal imaging: 1) MICCAI Challenge 2021 FLARE, 2) MICCAI Challenge 2021 FeTA, and 3) MICCAI Challenge 2022 AMOS. 3D UX-Net consistently outperforms SwinUNETR with improvement from 0.929 to 0.938 Dice (FLARE2021) and 0.867 to 0.874 Dice (Feta2021). We further evaluate the transfer learning capability of 3D UX-Net with AMOS2022 and demonstrates another improvement of $2.27\%$ Dice (from 0.880 to 0.900). The source code with our proposed model are available at https://github.com/MASILab/3DUX-Net. | We propose a lightweight network 3D UX-Net that simulates hierarchical transformer behavior with large kernel depthwise convolution and introduce pointwise depthwise scaling to scale features with less model parameters for volumetric segmentation. |
Aligning the visual and language spaces requires to train deep neural networks from scratch on giant multimodal datasets; CLIP trains both an image and a text encoder, while LiT manages to train just the latter by taking advantage of a pretrained vision network. In this paper, we show that sparse relative representations are sufficient to align text and images without training any network. Our method relies on readily available single-domain encoders (trained with or without supervision) and a modest (in comparison) number of image-text pairs. ASIF redefines what constitutes a multimodal model by explicitly disentangling memory from processing: here the model is defined by the embedded pairs of all the entries in the multimodal dataset, in addition to the parameters of the two encoders. Experiments on standard zero-shot visual benchmarks demonstrate the typical transfer ability of image-text models. Overall, our method represents a simple yet surprisingly strong baseline for foundation multi-modal models, raising important questions on their data efficiency and on the role of retrieval in machine learning. | How to build a CLIP-like model with two pretrained encoders and a limited amount of image-text pairs without tuning a neuron. |
Neural networks often exhibit emergent behavior in which qualitatively new capabilities that arise from scaling up the number of parameters, training data, or even the number of steps. One approach to understanding emergence is to find the continuous \textit{progress measures} that underlie the seemingly discontinuous qualitative changes. In this work, we argue that progress measures can be found via mechanistic interpretability---that is, by reverse engineering learned models into components and measuring the progress of each component over the course of training. As a case study, we study small transformers trained on a modular arithmetic tasks with emergent grokking behavior. We fully reverse engineer the algorithm learned by these networks, which uses discrete fourier transforms and trigonometric identities to convert addition to rotation about a circle. After confirming the algorithm via ablation, we then use our understanding of the algorithm to define progress measures that precede the grokking phase transition on this task. We see our result as demonstrating both that it is possible to fully reverse engineer trained networks, and that doing so can be invaluable to understanding their training dynamics. | We fully reverse engineer how one-layer transformers implement modular addition, and use this knowledge to explain grokking. |
In non-asymptotic statistical inferences, variance-type parameters of sub-Gaussian distributions play a crucial role. However, direct estimation of these parameters based on the empirical moment generating function (MGF) is infeasible. To this end, we recommend using a sub-Gaussian intrinsic moment norm [Buldygin and Kozachenko (2000), Theorem 1.3] through maximizing a series of normalized moments. Importantly, the recommended norm can not only recover the exponential moment bounds for the corresponding MGFs, but also lead to tighter Hoeffiding's sub-Gaussian concentration inequalities. In practice, intrinsic moment norm can be robustly and consistently estimated via a simple plug-in approach. Our theoretical results are applied to non-asymptotic analysis, including the multi-armed bandit. | Tight Non-asymptotic Inference |
Graph neural networks (GNNs) are prominent in the graph machine learning domain, owing to their strong performance across various tasks. A recent focal area is the space of graph self-supervised learning (SSL), which aims to derive useful node representations without labeled data. Notably, many state-of-the-art graph SSL methods are contrastive methods, which use a combination of positive and negative samples to learn node representations. Owing to challenges in negative sampling (slowness and model sensitivity), recent literature introduced non-contrastive methods, which instead only use positive samples. Though such methods have shown promising performance in node-level tasks, their suitability for link prediction tasks, which are concerned with predicting link existence between pairs of nodes (and have broad applicability to recommendation systems contexts) is yet unexplored. In this work, we extensively evaluate the performance of existing non-contrastive methods for link prediction in both transductive and inductive settings. While most existing non-contrastive methods perform poorly overall, we find that, surprisingly, BGRL generally performs well in transductive settings. However, it performs poorly in the more realistic inductive settings where the model has to generalize to links to/from unseen nodes. We find that non-contrastive models tend to overfit to the training graph and use this analysis to propose T-BGRL, a novel non-contrastive framework that incorporates cheap corruptions to improve the generalization ability of the model. This simple modification strongly improves inductive performance in 5/6 of our datasets, with up to a 120% improvement in Hits@50 - all with comparable speed to other non-contrastive baselines, and up to $14\times$ faster than the best-performing contrastive baseline. Our work imparts interesting findings about non-contrastive learning for link prediction and paves the way for future researchers to further expand upon this area. | We evaluate the performance of non-contrastive methods on link prediction and propose a new method to improve its performance in the inductive setting. |
Dot-product is a central building block in neural networks.
However, multiplication ($\texttt{mult}$) in dot-product consumes intensive energy and space costs that challenge deployment on resource-constrained edge devices.
In this study, we realize energy-efficient neural networks by exploiting a $\texttt{mult}$-less, sparse dot-product. We first reformulate a dot-product between an integer weight and activation into an equivalent operation comprised of additions followed by bit-shifts ($\texttt{add-shift-add}$).
In this formulation, the number of $\texttt{add}$ operations equals the number of bits of the integer weight in binary format.
Leveraging this observation, we propose Bit-Pruning, which removes unnecessary bits in each weight value during training to reduce the energy consumption of $\texttt{add-shift-add}$. Bit-Pruning can be seen as soft Weight-Pruning as it prunes bits, not the whole weight element.
In extensive experiments, we demonstrate that sparse $\texttt{mult}$-less networks trained with Bit-Pruning show a better accuracy-energy trade-off than sparse $\texttt{mult}$ networks trained with Weight-Pruning. | Mult-less dot-product comprised of add and shift; add is pruned during training to reduce energy |
Graph-level anomaly detection aims at depicting anomalous individual graphs in a graph set. Due to its significance in various real-world application fields, such as identifying rare molecules in chemistry and detecting potential frauds in online social networks, graph-level anomaly detection has received great attention. In distinction from node- and edge-level anomaly detection that is devoted to identifying anomalies on a single graph, graph-level anomaly detection faces more significant challenges because both the intra- and inter-graph structural and attribute patterns need to be taken into account to distinguish anomalies that exhibit deviating structures, rare attributes or the both. Although deep graph representation learning shows effectiveness in fusing high-level representations and capturing characters of individual graphs, most of the existing works are defective in graph-level anomaly detection because of their limited capability in exploring information across graphs, the imbalanced data distribution of anomalies, and low interpretability of the black-box graph neural networks (GNNs). To bridge these gaps, we propose a novel deep evolutionary graph mapping framework named GmapAD, which can adaptively map each graph into a new feature space based on its similarity to a set of representative nodes chosen from the graph set. By automatically adjusting the candidate nodes using a specially designed evolutionary algorithm, anomalies and normal graphs are mapped to separate areas in the new feature space where a clear boundary between them can be learned. The selected candidate nodes can therefore be regarded as a benchmark for explaining anomalies because anomalies are more dissimilar/similar to the benchmark than normal graphs. Through our extensive experiments on nine real-world datasets, we demonstrate that exploring both intra- and inter-graph structural and attribute information are critical to spot anomalous graphs, and our framework outperforms the state of the art on all datasets used in the experiments. | We propose a novel graph-level anomaly detection framework by mapping graphs into a specially designed feature space in which anomalies and normal graphs are well-separated. |
While kernel regression remains an important practical method, its connection to neural networks as their width becomes large has initiated fresh research. These neural kernels have drastically increased performance on diverse and nonstandard data modalities but require significantly more compute, which previously limited their application to smaller datasets. We address this by massively parallelizing their computation across many GPUs. We combine this with a distributed, preconditioned conjugate gradients algorithm to enable kernel regression at a large scale (i.e. up to 5 million examples). Using this approach, we study scaling laws of several neural kernels across many orders of magnitude for the CIFAR-5m dataset. Using data augmentation to expand the original CIFAR-10 training dataset by a factor of 20, we obtain a test accuracy of 91.2\% (SotA for a pure kernel method). Finally, we explore other data modalities, obtaining results on protein and small molecule prediction tasks that are competitive with SotA methods.
| We enable kernel regression with infinite-width neural networks at a larger scale than was previously possible to calculate scaling laws across many orders of magnitude and achieve SotA results on protein and small molecule prediction benchmarks. |
When learning task-oriented dialogue (ToD) agents, reinforcement learning (RL) techniques can naturally be utilized to train dialogue strategies to achieve user-specific goals. Prior works mainly focus on adopting advanced RL techniques to train the ToD agents, while the design of the reward function is not well studied. This paper aims at answering the question of how to efficiently learn and leverage a reward function for training end-to-end (E2E) ToD agents. Specifically, we introduce two generalized objectives for reward-function learning, inspired by the classical learning-to-rank literature. Further, we utilize the learned reward function to guide the training of the E2E ToD agent. With the proposed techniques, we achieve competitive results on the E2E response-generation task on the Multiwoz 2.0 dataset. Source code and checkpoints are publicly released at https://github.com/Shentao-YANG/Fantastic_Reward_ICLR2023. | we propose techniques for learning and utilizing reward functions that can be used for training task-oriented dialogue agents |
Existing inverse reinforcement learning methods (e.g. MaxEntIRL, $f$-IRL) search over candidate reward functions and solve a reinforcement learning problem in the inner loop. This creates a rather strange inversion where a harder problem, reinforcement learning, is in the inner loop of a presumably easier problem, imitation learning. In this work, we show that better utilization of expert demonstrations can reduce the need for hard exploration in the inner RL loop, hence accelerating learning. Specifically, we propose two simple recipes: (1) placing expert transitions into the replay buffer of the inner RL algorithm (e.g. Soft-Actor Critic) which directly informs the learner about high reward states instead of forcing the learner to discover them through extensive exploration, and (2) using expert actions in Q value bootstrapping in order to improve the target Q value estimates and more accurately describe high value expert states. Our methods show significant gains over a MaxEntIRL baseline on the benchmark MuJoCo suite of tasks, speeding up recovery to 70\% of deterministic expert performance by 2.13x on HalfCheetah-v2, 2.6x on Ant-v2, 18x on Hopper-v2, and 3.36x on Walker2d-v2. | This paper presents two simple, general-purpose recipes for accelerating inverse reinforcement learning through better utilization of expert information. |
Feature selection is of great importance and applies in lots of fields, such as medical and commercial. Wrapper methods, directly comparing the performance of different feature combinations, are widely used in real-world applications. However, selecting effective features meets the following two main challenges: 1) feature combinations are distributed in a huge discrete space; and 2) efficient and precise combinations evaluation is hard. To tackle these challenges, we propose a novel deep meta-learning-based feature selection framework, termed MetaFS, containing a Feature Subset Sampler (FSS) and a Meta Feature Estimator (MetaFE), which transforms the discrete search space into continuous and adopts meta-learning technique for effective feature selection. Specifically, FSS parameterizes the distribution of discrete search space and applies gradient-based methods to optimize. MetaFE learns the representations of different feature combinations, and dynamically generates unique models without retraining for efficient and precise combination evaluation. We adopt a bi-level optimization strategy to optimize the MetaFS. After optimization, we evaluate multiple feature combinations sampled from the converged distribution (i.e., the condensed search space) and select the optimal one. Finally, we conduct extensive experiments on two datasets, illustrating the superiority of MetaFS over 7 state-of-the-art methods. | We propose a meta-learning-based wrapper feature selection framewrok that doesn't require re-training plenty of models to evaluate different subsets. |
Concept Bottleneck Models (CBMs) map the inputs onto a set of interpretable concepts (``the bottleneck'') and use the concepts to make predictions. A concept bottleneck enhances interpretability since it can be investigated to understand what concepts the model "sees" in an input and which of these concepts are deemed important. However, CBMs are restrictive in practice as they require dense concept annotations in the training data to learn the bottleneck. Moreover, CBMs often do not match the accuracy of an unrestricted neural network, reducing the incentive to deploy them in practice. In this work, we address these limitations of CBMs by introducing Post-hoc Concept Bottleneck models (PCBMs). We show that we can turn any neural network into a PCBM without sacrificing model performance while still retaining the interpretability benefits. When concept annotations are not available on the training data, we show that PCBM can transfer concepts from other datasets or from natural language descriptions of concepts via multimodal models. A key benefit of PCBM is that it enables users to quickly debug and update the model to reduce spurious correlations and improve generalization to new distributions. PCBM allows for global model edits, which can be more efficient than previous works on local interventions that fix a specific prediction. Through a model-editing user study, we show that editing PCBMs via concept-level feedback can provide significant performance gains without using data from the target domain or model retraining. | We present a method to turn any neural network into a concept bottleneck model without sacrificing model performance, retaining interpretability benefits along with easy model editing. |
Animal navigation research posits that organisms build and maintain internal spa- tial representations, or maps, of their environment. We ask if machines – specifically, artificial intelligence (AI) navigation agents – also build implicit (or ‘mental’) maps. A positive answer to this question would (a) explain the surprising phenomenon in recent literature of ostensibly map-free neural-networks achieving strong performance, and (b) strengthen the evidence of mapping as a fundamental mechanism for navigation by intelligent embodied agents, whether they be biological or artificial. Unlike animal navigation, we can judiciously design the agent’s perceptual system and control the learning paradigm to nullify alternative navigation mechanisms. Specifically, we train ‘blind’ agents – with sensing limited to only egomotion and no other sensing of any kind – to perform PointGoal navigation (‘go to $\Delta$x, $\Delta$y’) via reinforcement learning. Our agents are composed of navigation-agnostic components (fully-connected and recurrent neural networks), and our experimental setup provides no inductive bias towards mapping. Despite these harsh conditions, we find that blind agents are (1) surprisingly effective navigators in new environments (∼95% success); (2) they utilize memory over long horizons (remembering ∼1,000 steps of past experience in an episode); (3) this memory enables them to exhibit intelligent behavior (following walls, detecting collisions, taking shortcuts); (4) there is emergence of maps and collision detection neurons in the representations of the environment built by a blind agent as it navigates; and (5) the emergent maps are selective and task dependent (e.g. the agent ‘forgets’ exploratory detours). Overall, this paper presents no new techniques for the AI audience, but a surprising finding, an insight, and an explanation. | ‘Blind’ AI navigation agents (with only egomotion sensing) can learn to navigate new environments and build map-like representations (supporting the ability to take shortcuts, follow walls, and predict free-space and collisions) of their environment. |
Existing models for learning representations in supervised classification problems are permutation invariant with respect to class labels. However, structured knowledge about the classes, such as hierarchical label structures, widely exists in many real-world datasets, e.g., the ImageNet and CIFAR benchmarks. How to learn representations that can preserve such structures among the classes remains an open problem. To approach this problem, given a tree of class hierarchy, we first define a tree metric between any pair of nodes in the tree to be the length of the shortest path connecting them. We then provide a method to learn the hierarchical relationship of class labels by approximately embedding the tree metric in the Euclidean space of features. More concretely, during supervised training, we propose to use the Cophenetic Correlation Coefficient (CPCC) as a regularizer for the cross-entropy loss to correlate the tree metric of classes and the Euclidean distance in the class-conditioned representations. Our proposed regularizer is computationally lightweight and easy to implement. Empirically, we demonstrate that this approach can help to learn more interpretable representations due to the preservation of the tree metric, and leads to better in-distribution generalization as well as under sub-population shifts over six real-world datasets. | We propose to learn structured representations that preserve the hierarchy between label classes by using CPCC as a regularizer. |
In recommender systems, users always choose the favorite items to rate, which leads to data missing not at random and poses a great challenge for unbiased evaluation and learning of prediction models. Currently, the doubly robust (DR) methods have been widely studied and demonstrate superior performance. However, in this paper, we show that DR methods are unstable and have unbounded bias, variance, and generalization bounds to extremely small propensities. Moreover, the fact that DR relies more on extrapolation will lead to suboptimal performance. To address the above limitations while retaining double robustness, we propose a stabilized doubly robust (StableDR) learning approach with a weaker reliance on extrapolation. Theoretical analysis shows that StableDR has bounded bias, variance, and generalization error bound simultaneously under inaccurate imputed errors and arbitrarily small propensities. In addition, we propose a novel learning approach for StableDR that updates the imputation, propensity, and prediction models cyclically, achieving more stable and accurate predictions. Extensive experiments show that our approaches significantly outperform the existing methods. | This paper proposes a theoretically guaranteed stabilized doubly robust learning approach that overcomes the shortcomings due to the presence of extremely small propensities in debiased recommendations. |
Most approaches for self-supervised learning (SSL) are optimised on curated balanced datasets, e.g. ImageNet, despite the fact that natural data usually exhibits long-tail distributions. In this paper, we analyse the behaviour of one of the most popular variants of SSL, i.e. contrastive methods, on imbalanced data. In particular, we investigate the role of the temperature parameter $\tau$ in the contrastive loss, by analysing the loss through the lens of average distance maximisation, and find that a large $\tau$ emphasises group-wise discrimination, whereas a small $\tau$ leads to a higher degree of instance discrimination. While $\tau$ has thus far been treated exclusively as a constant hyperparameter, in this work, we propose to employ a dynamic $\tau$ and show that a simple cosine schedule can yield significant improvements in the learnt representations. Such a schedule results in a constant `task switching' between an emphasis on instance discrimination and group-wise discrimination and thereby ensures that the model learns both group-wise features, as well as instance-specific details. Since frequent classes benefit from the former, while infrequent classes require the latter, we find this method to consistently improve separation between the classes in long-tail data without any additional computational cost. | Simple temperature schedules in self-supervised contrastive learning improve representation learning on long-tail distributions |
Understanding spatiotemporal relationships among several agents is of considerable relevance for many domains. Team sports represent a particularly interesting real-world proving ground since modeling interacting athletes requires capturing highly dynamic and complex agent-agent dependencies in addition to temporal components. However, existing generative methods in this field either entangle all latent factors into a single variable and are thus constrained in practical applicability, or they focus on uncovering interaction structures, which restricts their generative ability. To address this gap, we propose a framework for multiagent trajectories that augments sequential generative models with latent social structures. First, we derive a novel objective via approximate inference using a disentangled latent space that accurately describes the data generating process in such systems. Based on the proposed training criterion, we then present a model architecture that unifies insights from neural interaction inference and graph-structured variational recurrent neural networks for generating collective movements while allocating latent information. We validate our model on data from professional soccer and basketball. Our framework not only improves upon existing state-of-the-art approaches in forecasting trajectories, but also infers semantically meaningful representations that can be used in downstream tasks. | We propose a novel framework for multiagent trajectories that augments sequential generative models with latent social structures. |
Mixup is an efficient data augmentation approach that improves the generalization of neural networks by smoothing the decision boundary with mixed data. Recently, dynamic mixup methods have improved previous static policies effectively (e.g., linear interpolation) by maximizing salient regions or maintaining the target in mixed samples. The discrepancy is that the generated mixed samples from dynamic policies are more instance discriminative than the static ones, e.g., the foreground objects are decoupled from the background. However, optimizing mixup policies with dynamic methods in input space is an expensive computation compared to static ones. Hence, we are trying to transfer the decoupling mechanism of dynamic methods from the data level to the objective function level and propose the general decoupled mixup (DM) loss. The primary effect is that DM can adaptively focus on discriminative features without losing the original smoothness of the mixup while avoiding heavy computational overhead. As a result, DM enables static mixup methods to achieve comparable or even exceed the performance of dynamic methods. This also leads to an interesting objective design problem for mixup training that we need to focus on both smoothing the decision boundaries and identifying discriminative features. Extensive experiments on supervised and semi-supervised learning benchmarks across seven classification datasets validate the effectiveness of DM by equipping it with various mixup methods. | This paper proposes a decoupled mixup (DM) loss that can adaptively mine discriminative features without losing smoothness and improve various mixup methods in a plug-and-play manner. |
Deep learning requires increasingly bigger models and datasets to improve generalization on unseen data, where some training data samples may be more informative than others. We investigate this assumption in supervised image classification by biasing SGD (Stochastic Gradient Descent) to sample important samples more often during training of a classifier. In contrast to state-of-the-art, our approach does not require additional training iterations to estimate the sample importance, because it computes estimates once during training using the training prediction probabilities. In experiments, we see that our learning technique converges on par or faster in terms of training iterations and can achieve higher test accuracy compared to state-of-the-art, especially when datasets are not suitably balanced. Results suggest that sample importance has intrinsic balancing properties and that an importance weighted class distribution can converge faster than the usual balanced class distribution. Finally, in contrast to recent work, we find that sample importance is model dependent. Therefore, calculating sample importance during training, rather than in a pre-processing step, may be the only viable way to go. | Biasing SGD towards important samples after a short warm-up phase yields fast convergence, auto-balancing of classes, efficient use of augmentations, and shows that sample importance is model specific. |
Partitioning a set of elements into subsets of a priori unknown sizes is essential in many applications. These subset sizes are rarely explicitly learned - be it the cluster sizes in clustering applications or the number of shared versus independent generative latent factors in weakly-supervised learning. Probability distributions over correct combinations of subset sizes are non-differentiable due to hard constraints, which prohibit gradient-based optimization. In this work, we propose the differentiable hypergeometric distribution. The hypergeometric distribution models the probability of different group sizes based on their relative importance. We introduce reparameterizable gradients to learn the importance between groups and highlight the advantage of explicitly learning the size of subsets in two typical applications: weakly-supervised learning and clustering. In both applications, we outperform previous approaches, which rely on suboptimal heuristics to model the unknown size of groups. | We propose the differentiable hypergeometric distribution and show the advantage of explicitly learning subset sizes. |
Deep latent variable models have achieved significant empirical successes in model-based reinforcement learning (RL) due to their expressiveness in modeling complex transition dynamics. On the other hand, it remains unclear theoretically and empirically how latent variable models may facilitate learning, planning, and exploration to improve the sample efficiency of RL. In this paper, we provide a representation view of the latent variable models for state-action value functions, which allows both tractable variational learning algorithm and effective implementation of the optimism/pessimism principle in the face of uncertainty for exploration. In particular, we propose a computationally efficient planning algorithm with UCB exploration by incorporating kernel embeddings of latent variable models. Theoretically, we establish the sample complexity of the proposed approach in the online and offline settings. Empirically, we demonstrate superior performance over current state-of-the-art algorithms across various benchmarks. | We show how the latent variable model can be used for representation learning in reinforcement learning, that can have superior empirical performance as well as complete sample complexity analysis. |
Value-based algorithms have achieved great successes in solving Reinforcement Learning problems via minimizing the mean squared Bellman error (MSBE). Temporal-difference (TD) algorithms such as Q-learning and SARSA often use stochastic gradient descent based optimization approaches to estimate the value function parameters, but fail to quantify their uncertainties. In our work, under the Kalman filtering paradigm, we establish a novel and scalable sampling framework based on stochastic gradient Markov chain Monte Carlo, which allows us to efficiently generate samples from the posterior distribution of deep neural network parameters. For TD-learning with both linear and nonlinear function approximation, we prove that the proposed algorithm converges to a stationary distribution, which allows us to measure uncertainties of the value function and its parameters. | We propose an efficient and scalable sampling framework for reinforcement learning, which enables uncertainty quantification and addresses the local trap issue suffered by the existing training approaches. |
Transfer learning is crucial in training deep neural networks on new target tasks. Current transfer learning methods always assume at least one of (i) source and target task label spaces overlap, (ii) source datasets are available, and (iii) target network architectures are consistent with source ones. However, holding these assumptions is difficult in practical settings because the target task rarely has the same labels as the source task, the source dataset access is restricted due to storage costs and privacy, and the target architecture is often specialized to each task. To transfer source knowledge without these assumptions, we propose a transfer learning method that uses deep generative models and is composed of the following two stages: pseudo pre-training (PP) and pseudo semi-supervised learning (P-SSL). PP trains a target architecture with an artificial dataset synthesized by using conditional source generative models. P-SSL applies SSL algorithms to labeled target data and unlabeled pseudo samples, which are generated by cascading the source classifier and generative models to condition them with target samples. Our experimental results indicate that our method can outperform the baselines of scratch training and knowledge distillation. | We propose a novel transfer learning method using conditional generative models pre-trained on source dataset for an inductive transfer learning setting where NN architectures are not consistent. |
The expressive power of Graph Neural Networks (GNNs) has been studied extensively through the lens of the Weisfeiler-Leman (WL) graph isomorphism test. Yet, many graphs arising in real-world applications come embedded in Euclidean space with an additional notion of geometric isomorphism, which is not covered by the WL framework. In this work, we propose a geometric version of the WL test (GWL) for discriminating geometric graphs while respecting the underlying physical symmetries: permutation, rotation, reflection, and translation. We use GWL to characterise the expressive power of GNNs that are invariant or equivariant to physical symmetries by studying the classes of geometric graphs that can or cannot be distinguished by these architectures. This allows us to formalise the advantages equivariant GNN layers have over their invariant counterparts in the Geometric Deep Learning blueprint. Finally, we connect our discrimination-based perspective with the universal approximation properties of geometric GNNs and prove they are two sides of the same coin. | We propose a geometric version of the Weisfeler-Leman graph isomorphism test to characterise the expressive power of GNNs for geometric graphs. |
While large language models (LLMs) like GPT-3 have achieved impressive results on multiple choice question answering (MCQA) tasks in the zero, one, and few-shot settings, they generally lag behind the MCQA state of the art (SOTA). MCQA tasks have traditionally been presented to LLMs like cloze tasks. An LLM is conditioned on a question (without the associated answer options) and its chosen option is the one assigned the highest probability after normalization (for length, etc.). A more natural prompting approach is to present the question and answer options to the LLM jointly and have it output the symbol (e.g., “A”) associated with its chosen answer option. This approach allows the model to explicitly compare answer options, reduces computational costs, and mitigates the effects of tokenization scheme and answer option representations on answer selection. For the natural approach to be effective, the LLM it is used with must be able to associate answer options with the symbols that represent them. The LLM needs what we term multiple choice symbol binding (MCSB) ability. This ability varies greatly by model. We show that a model with high MCSB ability performs much better with the natural approach than with the traditional approach across 20 diverse datasets and largely closes the gap with the SOTA, suggesting that the MCQA ability of LLMs has been previously underestimated. | Large language models that can effectively associate multiple choice answer options with symbols can be prompted in a way that yields dramatically improved performance on multiple choice question answering tasks. |
Online meta-learning has recently emerged as a marriage between batch meta-learning and online learning, for achieving the capability of quick adaptation on new tasks in a lifelong manner. However, most existing approaches focus on the restrictive setting where the distribution of the online tasks remains fixed with known task boundaries. In this work we relax these assumptions and propose a novel algorithm for task-agnostic online meta-learning in non-stationary environments. More specifically, we first propose two simple but effective detection mechanisms of task switches and distribution shift based on empirical observations, which serve as a key building block for more elegant online model updates in our algorithm: the task switch detection mechanism allows reusing of the best model available for the current task at hand, and the distribution shift detection mechanism differentiates the meta model update so as to preserve the knowledge for in-distribution tasks and quickly learn the new knowledge for out-of-distribution tasks.
Motivated by the recent advance in online learning, our online meta model updates are based only on the current data, which eliminates the need of storing previous data as required in most existing methods. This crucial choice is also well supported by our theoretical analysis of dynamic regret in online meta-learning, where a sublinear regret can be achieved by updating the meta model at each round using the current data only. Empirical studies on three different benchmarks clearly demonstrate the significant advantage of our algorithm over related baseline approaches. | We propose a novel algorithm for task-agnostic online meta-learning in non-stationary environments without knowledge of task boundaries. |
Machine learning models are often developed in a way that prioritizes task-specific performance but defers the understanding of how they actually work. This is especially true nowadays for deep neural networks. In this paper, we step back and consider the basic problem of understanding a learned model represented as a smooth scalar-valued function. We introduce HeatFlow, a framework based upon the heat diffusion process for interpreting the multi-scale behavior of the model around a test point. At its core, our approach looks into the heat flow initialized at the function of interest, which generates a family of functions with increasing smoothness. By applying differential operators to these smoothed functions, summary statistics (i.e., explanations) characterizing the original model on different scales can be drawn. We place an emphasis on studying the heat flow on data manifold, where the model is trained and expected to be well behaved. Numeric approximation procedures for implementing the proposed method in practice are discussed and demonstrated on image recognition tasks. | Solve the heat equation to interpret neural networks. |
Offline reinforcement learning (RL) is a challenging setting where existing off-policy actor-critic methods perform poorly due to overestimating of out-of-distribution state-action pairs. Thus, various additional augmentations are proposed to keep the learned policy close to the offline dataset (or the behavior policy). In this work, starting from the analysis of offline monotonic policy improvement, we reach a surprising conclusion that online on-policy algorithms are naturally able to solve offline RL. Specifically, the inherent conservatism of these on-policy algorithms is exactly what the offline RL method needs to overcome the overestimation. Based on this, we propose Behavior Proximal Policy Optimization (BPPO), which solves offline RL without any extra constraint or regularization introduced compared to PPO. Extensive experiments on the D4RL benchmark empirically show this extremely succinct method outperforms state-of-the-art offline RL algorithms. Our implementation is available at https://github.com/Dragon-Zhuang/BPPO. | We propose Behavior Proximal Policy Optimization (BPPO), which bases on on-policy method (PPO) and effectively solves offline RL without any extra constraint or regularization introduced. |
In machine learning, we traditionally evaluate the performance of a single model, averaged over a collection of test inputs. In this work, we propose a new approach: we measure the performance of a collection of models when evaluated at *single input point*. Specifically, we study a point's *profile*: the relationship between models' average performance on the test distribution and their pointwise performance on this individual point. We find that profiles can yield new insights into the structure of both models and data---in and out-of-distribution. For example, we empirically show that real data distributions consist of points with qualitatively different profiles. On one hand, there are ``compatible'' points with strong correlation between the pointwise and average performance. On the other hand, there are points with weak and even *negative* correlation: cases where improving overall model accuracy actually *hurts* performance on these inputs. As an application, we use profiles to construct a dataset we call CIFAR-10-NEG: a subset of CINIC-10 such that for standard models, accuracy on CIFAR-10-NEG is *negatively correlated* with CIFAR-10 accuracy. Illustrating for the first time an OOD dataset that completely inverts ``accuracy-on-the-line'' (Miller et al., 2021). | We propose a new lens for studying the pointwise performance of learning algorithms which reveals new insights into their behavior and goes beyond traditional notions of in-distribution and "out-of-distribution" learning. |
We explore the value-based method and policy gradient combination in multi-agent reinforcement learning (MARL). In value-based MARL, {\itshape{Individual-Global-Max}} (IGM) principle plays an important role, which maintains the consistency between joint and local action values. At the same time, IGM is difficult to guarantee in multi-agent policy gradient methods due to stochastic exploration and conflicting gradient directions. In this paper, we propose a novel multi-agent policy gradient algorithm called {\itshape{Advantage Constrained Proximal Policy Optimization}} (ACPPO). Based on {\itshape{multi-agent advantage decomposition lemma}}, ACPPO introduces an advantage network for each agent to estimate current local state-action advantage. The coefficient of each agent constrains the joint-action advantage according to the consistency of the estimated joint-action advantage and local advantage. Unlike previous policy gradient-based MARL algorithms, ACPPO does not need an extra sampled baseline to reduce variance. We evaluate the proposed methods for continuous matrix game and Multi-Agent MuJoCo tasks. Results show that ACPPO outperforms the baselines such as MAPPO, MADDPG, and HAPPO. | A multi-agent policy gradient reinforcement learning based on local advantage constraned. |
Multi-document summarization entails producing concise synopses of collections of inputs. For some applications, the synopsis should accurately \emph{synthesize} inputs with respect to a key property or aspect. For example, a synopsis of film reviews all written about a particular movie should reflect the average critic consensus. As a more consequential example, consider narrative summaries that accompany biomedical \emph{systematic reviews} of clinical trial results. These narratives should fairly summarize the potentially conflicting results from individual trials.
In this paper we ask: To what extent do modern multi-document summarization models implicitly perform this type of synthesis? To assess this we perform a suite of experiments that probe the degree to which conditional generation models trained for summarization using standard methods yield outputs that appropriately synthesize inputs. We find that existing models do partially perform synthesis, but do so imperfectly. In particular, they are over-sensitive to changes in input ordering and under-sensitive to changes in input compositions (e.g., the ratio of positive to negative movie reviews). We propose a simple, general method for improving model synthesis capabilities by generating an explicitly diverse set of candidate outputs, and then selecting from these the string best aligned with the expected aggregate measure for the inputs, or \emph{abstaining} when the model produces no good candidate. This approach improves model synthesis performance. Our hope is that by highlighting the need for synthesis (in some summarization settings), this work motivates further research into multi-document summarization methods and learning objectives that explicitly account for the need to synthesize. | We measure if multidocument summarization models can effectively synthesize contrasting inputs, and explore methods to change synthesis performance. |
Modern deep learning systems are increasingly deployed in situations such as personalization and federated learning where it is necessary to support i) learning on small amounts of data, and ii) communication efficient distributed training protocols. In this work, we develop FiLM Transfer (FiT) which fulfills these requirements in the image classification setting by combining ideas from transfer learning (fixed pretrained backbones and fine-tuned FiLM adapter layers) and meta-learning (automatically configured Naive Bayes classifiers and episodic training) to yield parameter efficient models with superior classification accuracy at low-shot. The resulting parameter efficiency is key for enabling few-shot learning, inexpensive model updates for personalization, and communication efficient federated learning. We experiment with FiT on a wide range of downstream datasets and show that it achieves better classification accuracy than the leading Big Transfer (BiT) algorithm at low-shot and achieves state-of-the art accuracy on the challenging VTAB-1k benchmark, with fewer than 1% of the updateable parameters. Finally, we demonstrate the parameter efficiency and superior accuracy of FiT in distributed low-shot applications including model personalization and federated learning where model update size is an important performance metric. | We propose FiT, a parameter efficient few-shot image classification system that uses a Naive Bayes head, FiLM layers that modulate a pretrained backbone, and an episodic fine-tuning protocol that achieves SOTA on the VTAB-1k benchmark. |
Differentially Private methods for training Deep Neural Networks (DNNs) have progressed recently, in particular with the use of massive batches and aggregated data augmentations for a large number of steps. These techniques require much more compute than their non-private counterparts, shifting the traditional privacy-accuracy trade-off to a privacy-accuracy-compute trade-off and making hyper-parameter search virtually impossible for realistic scenarios. In this work, we decouple privacy analysis and experimental behavior of noisy training to explore the trade-off with minimal computational requirements. We first use the tools of Rényi Differential Privacy (RDP) to show that the privacy budget, when not overcharged, only depends on the total amount of noise (TAN) injected throughout training. We then derive scaling laws for training models with DP-SGD to optimize hyper-parameters with more than a $100\times$ reduction in computational budget. We apply the proposed method on CIFAR-10 and ImageNet and, in particular, strongly improve the state-of-the-art on ImageNet with a $+9$ points gain in accuracy for a privacy budget $\varepsilon=8$. | Computationally friendly hyper-parameter search with DP-SGD for new state-of-the-art performance on ImageNet. |
SGD (with momentum) and AdamW are the two most commonly used optimizers for fine-tuning large neural networks in computer vision. When the two methods perform the same, SGD is preferable because it uses less memory and is more efficient than AdamW. However, when evaluating on downstream tasks that differ significantly from pretraining, we find that across five popular benchmarks SGD fine-tuning gets substantially lower accuracies than AdamW on many modern vision models such as Vision Transformers and ConvNeXts---especially out-of-distribution (OOD). We find that such large gaps arise in instances where the fine-tuning gradients in the first (``embedding'') layer are much larger than the rest of the model. Our analysis suggests an easy fix: if we simply freeze the embedding layer (0.7\% of the parameters), SGD performs competitively with AdamW while using less memory across a suite of benchmarks. Our insights lead to state-of-the-art accuracies on popular distribution shift benchmarks: WILDS-FMoW, WILDS-Camelyon, BREEDS-Living-17, Waterbirds, and DomainNet. | SGD can do worse than AdamW under distribution shifts, but simple changes make SGD competitive |
Injustice occurs when someone experiences unfair treatment or their rights are violated. In the context of news media, injustices represent a form of bias through which discriminatory narratives can arise and spread. The automated identification of injustice in text has received little attention, due in part to the fact that underlying stereotypes are rarely explicitly stated and that instances often occur unconsciously due to the pervasive nature of prejudice in society. Here, we leverage the combined use of a fine-tuned BERT-based bias detection model, two stereotype detection models, and a lexicon-based approach to show that epistemological biases (i.e., words, which through their use, presupposes, entails, asserts, hedges, or boosts text to erode or assert a person's capacity as a knower) can assist with the automatic detection of injustice in text. | We leverage the combined use of a fine-tuned epistemological detection model, two stereotype detection models, and a lexicon-based approach to show that epistemological biases can assist with the automatic detection of injustices in text. |
Subsets and Splits