Datasets:

kreas

/

DAML4-TLDR-GEN-DATASET

like 0

Image restoration aims to reconstruct the latent sharp image from its corrupted counterpart. Besides dealing with this long-standing task in the spatial domain, a few approaches seek solutions in the frequency domain in consideration of the large discrepancy between spectra of sharp/degraded image pairs. However, these works commonly utilize transformation tools, e.g., wavelet transform, to split features into several frequency parts, which is not flexible enough to select the most informative frequency component to recover. In this paper, we exploit a multi-branch and content-aware module to decompose features into separate frequency subbands dynamically and locally, and then accentuate the useful ones via channel-wise attention weights. In addition, to handle large-scale degradation blurs, we propose an extremely simple decoupling and modulation module to enlarge the receptive field via global and window-based average pooling. Integrating two developed modules into a U-Net backbone, the proposed Selective Frequency Network (SFNet) performs favorably against state-of-the-art algorithms on five image restoration tasks, including single-image defocus deblurring, image dehazing, image motion deblurring, image desnowing, and image deraining.

We propose a novel network to recover the most useful frequency component for image restoration via frequency selection.

There has been a debate on the choice of plain vs. hierarchical vision transformers, where researchers often believe that the former (e.g., ViT) has a simpler design but the latter (e.g., Swin) enjoys higher recognition accuracy. Recently, the emerge of masked image modeling (MIM), a self-supervised visual pre-training method, raised a new challenge to vision transformers in terms of flexibility, i.e., part of image patches or tokens are to be discarded, which seems to claim the advantages of plain vision transformers. In this paper, we delve deep into the comparison between ViT and Swin, revealing that (i) the performance gain of Swin is mainly brought by a deepened backbone and relative positional encoding, (ii) the hierarchical design of Swin can be simplified into hierarchical patch embedding (proposed in this work), and (iii) other designs such as shifted-window attentions can be removed. By removing the unnecessary operations, we come up with a new architecture named HiViT (short for hierarchical ViT), which is simpler and more efficient than Swin yet further improves its performance on fully-supervised and self-supervised visual representation learning. In particular, after pre-trained using masked autoencoder (MAE) on ImageNet-1K, HiViT-B reports a 84.6% accuracy on ImageNet-1K classification, a 53.3% box AP on COCO detection, and a 52.8% mIoU on ADE20K segmentation, significantly surpassing the baseline. Code is available at https://github.com/zhangxiaosong18/hivit.

A novel hierarchical vision transformer that is stronger and faster when applied to masked image modeling

The Teacher-Student Framework (TSF) is a reinforcement learning setting where a teacher agent guards the training of a student agent by intervening and providing online demonstrations. Assuming optimal, the teacher policy has the perfect timing and capability to intervene in the learning process of the student agent, providing safety guarantee and exploration guidance. Nevertheless, in many real-world settings it is expensive or even impossible to obtain a well-performing teacher policy. In this work, we relax the assumption of a well-performing teacher and develop a new method that can incorporate arbitrary teacher policies with modest or inferior performance. We instantiate an Off-Policy Reinforcement Learning algorithm, termed Teacher-Student Shared Control (TS2C), which incorporates teacher intervention based on trajectory-based value estimation. Theoretical analysis validates that the proposed TS2C algorithm attains efficient exploration and substantial safety guarantee without being affected by the teacher's own performance. Experiments on various continuous control tasks show that our method can exploit teacher policies at different performance levels while maintaining a low training cost. Moreover, the student policy surpasses the imperfect teacher policy in terms of higher accumulated reward in held-out testing environments. Code is available at https://metadriverse.github.io/TS2C.

Introducing a new policy optimization method exploiting imperfect online demonstrations from a guardian policy.

The contextual bandit is a classic problem in sequential decision making under uncertainty that finds broad application to tasks in precision medicine, personalized education, and drug discovery. Here, a decision maker repeatedly receives a context, takes an action, and then observes an associated outcome, with the goal of choosing actions that achieve a minimal regret. However, in many settings, the context is not given, and the decision maker must instead collect some information to infer a context before proceeding. For example, when a doctor does not have prior information about a patient, they might ask a sequence of questions before recommending a medical treatment. In this paper, we aim to develop methods for this setting—which we refer to as the \emph{survey bandit}—where the decision maker is not given access to the context but can ask a finite sequence of questions to gain information about the context before taking an action and observing an outcome. Using insights from Bayesian optimal experimental design (BOED) and decision-theoretic information theory, we view the interaction with each user as a BOED task, where the goal is to ask a sequence of questions that elicit the most information about the optimal action for this user. Our procedure is agnostic to the choice of probabilistic model, and we demonstrate its usefulness in a few common classes of distributions. Our algorithm achieves significantly better performance on both synthetic and real data relative to existing baseline methods while remaining statistically efficient, interpretable, and computationally friendly.

We develop a Bayesian optimal experimental design procedure for the survey bandit setting.

In this work, we study offline preference-based reinforcement learning (PbRL), which relaxes the two fundamental supervisory signals in standard reinforcement learning (online accessible transition dynamics and rewards). In other words, the agent is provided with fixed offline trajectory transitions and human preferences between pairs of trajectories. Due to the orthogonality property of rewards and dynamics, one common practice is combining prior PbRL-based reward learning objectives with off-the-shelf offline RL algorithms to bridge preference modeling and offline learning. However, such two isolated optimizations require learning a separate reward function and thus place an information bottleneck on reward learning (the bridge). As an alternative, we propose offline preference-guided policy optimization (OPPO), an end-to-end offline PbRL formulation, which jointly learns to model the preference (for finding the optimal task policy) and the offline data (for eliminating OOD). In particular, OPPO introduces an offline hindsight information matching objective and a preference modeling objective. Then, iterating the two objectives over, we can directly extract a well-performing decision policy, avoiding a separate reward learning. We empirically show that OPPO can effectively model the offline preference and outperform prior competing baselines (including the offline RL algorithms performed over the true reward function).

We propose an end-to-end offline preference-based reinforcement learning formulation that directly optimizes the policy by preference supervision without learning a separate reward function.

Neural networks produced by standard training are known to suffer from poor accuracy on rare subgroups despite achieving high accuracy on average, due to the correlations between certain spurious features and labels. Previous approaches based on worst-group loss minimization (\textit{e.g.} Group-DRO) are effective in improving worse-group accuracy but require expensive group annotations for all the training samples. In this paper, we focus on the more challenging and realistic setting where group annotations are only available on a small validation set or are not available at all. We propose \bam, a novel two-stage training algorithm: in the first stage, the model is trained using a \emph{bias amplification} scheme via introducing a learnable \emph{auxiliary variable} for each training sample together with the adoption of squared loss; in the second stage, we upweight the samples that the bias-amplified model misclassifies, and then continue training the same model on the reweighted dataset. Empirically, \bam leads to consistent improvement over its counterparts in worst-group accuracy, resulting in state-of-the-art performance in spurious correlation benchmarks in computer vision and natural language processing. Moreover, we find a simple stopping criterion that completely removes the need for group annotations, with little or no loss in worst-group accuracy.

We propose a novel two-stage training algorithm that achieves the state-of-the-art worst-group accuracy on test data without group information.

Simplicial Embeddings (SEM) are representations learned through self-supervised learning (SSL), wherein a representation is projected into $L$ simplices of $V$ dimensions each using a \texttt{softmax} operation. This procedure conditions the representation onto a constrained space during pretraining and imparts an inductive bias for group sparsity. For downstream classification, we formally prove that the SEM representation leads to better generalization than an unnormalized representation. Furthermore, we empirically demonstrate that SSL methods trained with SEMs have improved generalization on natural image datasets such as CIFAR-100 and ImageNet. Finally, when used in a downstream classification task, we show that SEM features exhibit emergent semantic coherence where small groups of learned features are distinctly predictive of semantically-relevant classes.

We use softmax to embed representations in a collection of simplices in SSL models, which offers improved generalization properties for downstream classification.

The approximation power of neural networks makes it an ideal tool to learn optimal transport maps. However, existing methods are mostly based on the Kantorovich duality and require regularization and/or special network structures such as Input Convex Neural Networks (ICNN). In this paper, we propose a direct constraint optimization algorithm for the computation of optimal transport maps based on the Monge formulation. We solve this constraint optimization problem by using three different methods: the penalty method, the augmented Lagrangian method, and the alternating direction method of multipliers method (AMDD). We demonstrate a significant improvement in the accuracy of learned optimal transport maps on benchmarks. Moreover, we show that our methods reduce the regularization effects and accurately learn the target distributions at lower transport cost.

We integrates constraint optimization algorithm and neural networks for the computation of optimal transport maps based on the Monge formulation.

Causal learning has long concerned itself with the accurate recovery of underlying causal mechanisms. Such causal modelling enables better explanations of out-of-distribution data. Prior works on causal learning assume that the high-level causal variables are given. However, in machine learning tasks, one often operates on low-level data like image pixels or high-dimensional vectors. In such settings, the entire Structural Causal Model (SCM) -- structure, parameters, \textit{and} high-level causal variables -- is unobserved and needs to be learnt from low-level data. We treat this problem as Bayesian inference of the latent SCM, given low-level data. For linear Gaussian additive noise SCMs, we present a tractable approximate inference method which performs joint inference over the causal variables, structure and parameters of the latent SCM from random, known interventions. Experiments are performed on synthetic datasets and a causally generated image dataset to demonstrate the efficacy of our approach. We also perform image generation from unseen interventions, thereby verifying out of distribution generalization for the proposed causal model.

bayesian inference over latent structural causal models from low level data and random, known interventions for linear Gaussian additive noise SCMs. Such a model also performs image generation from unseen interventions,

Policy robustness in Reinforcement Learning (RL) may not be desirable at any price; the alterations caused by robustness requirements from otherwise optimal policies should be explainable and quantifiable. Policy gradient algorithms that have strong convergence guarantees are usually modified to obtain robust policies in ways that do not preserve algorithm guarantees, which defeats the purpose of formal robustness requirements. In this work we study a notion of robustness in partially observable MDPs where state observations are perturbed by a noise-induced stochastic kernel. We characterize the set of policies that are maximally robust by analysing how the policies are altered by this kernel. We then establish a connection between such robust policies and certain properties of the noise kernel, as well as with structural properties of the underlying MDPs, constructing sufficient conditions for policy robustness. We use these notions to propose a robustness-inducing scheme, applicable to any policy gradient algorithm, to formally trade off the reward achieved by a policy with its robustness level through lexicographic optimization, which preserves convergence properties of the original algorithm. We test the the proposed approach through numerical experiments on safety-critical RL environments, and show how the proposed method helps achieve high robustness when state errors are introduced in the policy roll-out.

We use lexicographic optimization to induce robustness in RL policies in a safe way.

We propose a Gaussian confidence measure and its optimization, for use in neural network classifiers. The measure comes with theoretical results, simultaneously resolving two pressing problems in NN classification: uncertainty quantification, and robustness. Existing research in uncertainty quantification mostly revolves around the confidence reflected in the input feature space. Instead, we focus on the learned representation of the network and analyze the confidence in the penultimate layer space. We formally prove that, independent of optimization-procedural effects, a set of centroids always exists such that softmax classifiers are nearest-centroid classifiers. Softmax confidence, however, does not reflect that the classification is based on nearest centroids: artificially inflated confidence is also given to out-of-distributions samples that are not near any centroid, but slightly less distant from one centroid than from the others. Our new confidence measure is centroid-based, and hence no longer suffers from the artificial confidence inflation of out-of-distribution samples. We also show that our proposed centroidal confidence measure is providing a robustness certificate against attacks. As such, it manages to reflect what the model doesn't know (as demanded by uncertainty quantification), and to resolve the issue of robustness of neural networks.

We address the overconfidence of neural networks and related issues with a new centroidal-based confidence measure.

The separation between training and deployment of machine learning models implies that not all scenarios encountered in deployment can be anticipated during training, and therefore relying solely on advancements in training has its limits. Out-of-distribution (OOD) detection is an important area that stress-tests a model’s ability to handle unseen situations: Do models know when they don’t know? Existing OOD detection methods either incur extra training steps, additional data or make nontrivial modifications to the trained network. In contrast, in this work, we propose an extremely simple, post-hoc, on-the-fly activation shaping method, ASH, where a large portion (e.g. 90%) of a sample’s activation at a late layer is removed, and the rest (e.g. 10%) simplified or lightly adjusted. The shaping is applied at inference time, and does not require any statistics calculated from training data. Experiments show that such a simple treatment enhances in-distribution and out- of-distribution sample distinction so as to allow state-of-the-art OOD detection on ImageNet, and does not noticeably deteriorate the in-distribution accuracy. Video, animation and code can be found at: https://andrijazz.github.io/ash.

We develop an extremely simple, post hoc, on-the-fly, and plug-and-play activation shaping method for out-of-distribution detection.

Model quantization is important for compressing models and improving computing speed. However, current researchers think that the loss function of quantizated model is usually higher than full precision model. This study provides a methodology for acquiring a mixed-precise quantization model with a lower loss than the full precision model. In addition, the analysis demonstrates that, throughout the inference process, the loss function is mostly affected by the noise of the layer inputs. In particular, we will demonstrate that neural networks with massive identity mappings are resistant to the quantization method. It is also difficult to improve the performance of these networks using quantization.

An appropriate quantisation method may reduce the neural network's loss, and we show our way in this paper.

We propose the Transformer with Multiresolution-head Attention (MrsFormer), a class of efficient transformers inspired by the multiresolution approximation (MRA) for approximating a signal f using wavelet bases. MRA decomposes a signal into components that lie on orthogonal subspaces at different scales. Similarly, MrsFormer decomposes the attention heads in the multi-head attention into fine-scale and coarse-scale heads, modeling the attention patterns between tokens and between groups of tokens. Computing the attention heads in MrsFormer requires significantly less computation and memory footprint compared to the standard softmax transformer with multi-head attention. We analyze and validate the advantage of MrsFormer over the standard transformers on a wide range of applications including image and time series classification.

We propose the Transformer with Multiresolution-head Attention (MrsFormer), a class of efficient transformers inspired by the multiresolution approximation (MRA) for approximating a signal f using wavelet bases

Large-scale diffusion models have achieved state-of-the-art results on text-to-image synthesis (T2I) tasks. Despite their ability to generate high-quality yet creative images, we observe that attribution-binding and compositional capabilities are still considered major challenging issues, especially when involving multiple objects. Attribute-binding requires the model to associate objects with the correct attribute descriptions, and compositional skills require the model to combine and generate multiple concepts into a single image. In this work, we improve these two aspects of T2I models to achieve more accurate image compositions. To do this, we incorporate linguistic structures with the diffusion guidance process based on the controllable properties of manipulating cross-attention layers in diffusion-based T2I models. We observe that keys and values in cross-attention layers have strong semantic meanings associated with object layouts and content. Therefore, by manipulating the cross-attention representations based on linguistic insights, we can better preserve the compositional semantics in the generated image. Built upon Stable Diffusion, a SOTA T2I model, our structured cross-attention design is efficient that requires no additional training samples. We achieve better compositional skills in qualitative and quantitative results, leading to a significant 5-8\% advantage in head-to-head user comparison studies. Lastly, we conduct an in-depth analysis to reveal potential causes of incorrect image compositions and justify the properties of cross-attention layers in the generation process.

We propose a training-free approach to incorporate language structured for compositional text-to-image synthesis

Adversarial patch attacks are an emerging security threat for real world deep learning applications. We present Demasked Smoothing, the first approach (up to our knowledge) to certify the robustness of semantic segmentation models against this threat model. Previous work on certifiably defending against patch attacks has mostly focused on image classification task and often required changes in the model architecture and additional training which is undesirable and computationally expensive. In Demasked Smoothing, any segmentation model can be applied without particular training, fine-tuning, or restriction of the architecture. Using different masking strategies, Demasked Smoothing can be applied both for certified detection and certified recovery. In extensive experiments we show that Demasked Smoothing can on average certify 63% of the pixel predictions for a 1% patch in the detection task and 46% against a 0.5% patch for the recovery task on the ADE20K dataset.

We propose the first approach for certified recovery and certified detection against adversarial patch attacks on semantic segmentation, which is based on novel masking schemes and image inpainting.

Information Extraction (IE) spans several tasks with different output structures, such as named entity recognition, relation extraction and event extraction. Previously, those tasks were solved with different models because of diverse task output structures. Through re-examining IE tasks, we find that all of them can be interpreted as extracting spans and span relations. We propose using the start and end token of a span to pinpoint the span in texts, and using the start-to-start and end-to-end token pairs of two spans to determine the relation. Hence, we can unify all IE tasks under the same token-pair classification formulation. Based on the reformulation, we propose a \textbf{U}nified \textbf{T}oken-pair \textbf{C}lassification architecture for \textbf{I}nformation \textbf{E}xtraction (\textbf{UTC-IE}), where we introduce Plusformer on top of the token-pair feature matrix. Specifically, it models axis-aware interaction with plus-shaped self-attention and local interaction with Convolutional Neural Network over token pairs. Experiments show that our approach outperforms task-specific and unified models on all tasks in 10 datasets, and achieves better or comparable results on 2 joint IE datasets. Moreover, UTC-IE speeds up over state-of-the-art models on IE tasks significantly in most datasets, which verifies the effectiveness of our architecture.

Reformulate and unify three IE tasks as token-pair classifications and propose Plusformer to effectively model the interaction between token pairs.

In text generation, models that generate text from scratch one token at a time are currently the dominant paradigm. Despite being performant, these models lack the ability to revise existing text, which limits their usability in many practical scenarios. We look to address this, with DiffusER (Diffusion via Edit-based Reconstruction), a new edit-based generative model for text based on denoising diffusion models -- a class of models that use a Markov chain of denoising steps to incrementally generate data. DiffusER is not only a strong generative model in general, rivalling autoregressive models on several tasks spanning machine translation, summarization, and style transfer; it can also perform other varieties of generation that standard autoregressive models are not well-suited for. For instance, we demonstrate that DiffusER makes it possible for a user to condition generation on a prototype, or an incomplete sequence, and continue revising based on previous edit steps.

We propose a generally applicable text generative model which takes inspiration from diffusion models and parameterises generation steps as text editing steps without compromising performance and adding flexibility.

Protein representation learning has primarily benefited from the remarkable development of language models (LMs). Accordingly, pre-trained protein models also suffer from a problem in LMs: a lack of factual knowledge. The recent solution models the relationships between protein and associated knowledge terms as the knowledge encoding objective. However, it fails to explore the relationships at a more granular level, i.e., the token level. To mitigate this, we propose Knowledge-exploited Auto-encoder for Protein (KeAP), which performs token-level knowledge graph exploration for protein representation learning. In practice, non-masked amino acids iteratively query the associated knowledge tokens to extract and integrate helpful information for restoring masked amino acids via attention. We show that KeAP can consistently outperform the previous counterpart on 9 representative downstream applications, sometimes surpassing it by large margins. These results suggest that KeAP provides an alternative yet effective way to perform knowledge enhanced protein representation learning.

We perform protein knowledge encoding by learning to exploit knowledge graphs for protein primary structure reasoning.

Neuroscience has provide many inspirations for the development of artificial intelligence, especially for neural networks for computer vision tasks. Recent research on animals' visual systems builds the connection between neural sparsity and animals' levels of evolution, based on which comparisons between two most influential vision architecture, Transformer and CNN, are carried out. In particular, the sparsity of attentions in Transformers is comprehensively studied, and previous knowledge on sparsity of neurons in CNNs is reviewed. In addition, a novel metric for neural sparsity is defined and ablation experiments are launched on various types of Transformer and CNN models. Finally, we draw the conclusion that more layers in models will result in higher sparsity, however, too many heads in Transformers may cause reduction of sparsity, which attributes to the significant overlap among effects of attention units.

Neural sparsity of Transformers and CNNs are defined and calculated, leading to striking conclusion.

A generative model based on a continuous-time normalizing flow between any pair of base and target probability densities is proposed. The velocity field of this flow is inferred from the probability current of a time-dependent density that interpolates between the base and the target in finite time. Unlike conventional normalizing flow inference methods based the maximum likelihood principle, which require costly backpropagation through ODE solvers, our interpolant approach leads to a simple quadratic loss for the velocity itself which is expressed in terms of expectations that are readily amenable to empirical estimation. The flow can be used to generate samples from either the base or target, and to estimate the likelihood at any time along the interpolant. In addition, the flow can be optimized to minimize the path length of the interpolant density, thereby paving the way for building optimal transport maps. In situations where the base is a Gaussian density, we also show that the velocity of our normalizing flow can also be used to construct a diffusion model to sample the target as well as estimate its score. However, our approach shows that we can bypass this diffusion completely and work at the level of the probability flow with greater simplicity, opening an avenue for methods based solely on ordinary differential equations as an alternative to those based on stochastic differential equations. Benchmarking on density estimation tasks illustrates that the learned flow can match and surpass conventional continuous flows at a fraction of the cost, and compares well with diffusions on image generation on CIFAR-10 and ImageNet $32 \times 32$. The method scales ab-initio ODE flows to previously unreachable image resolutions, demonstrated up to $128\times128$.

A method is proposed to construct normalizing flows based on stochastic interpolants, yielding an efficient training algorithm compared to equivalent ODE methods, and providing a theoretical framework to map score based diffusions to ODEs.

Many works show that node-level predictions of Graph Neural Networks (GNNs) are unrobust to small, often termed adversarial, changes to the graph structure. However, because manual inspection of a graph is difficult, it is unclear if the studied perturbations always preserve a core assumption of adversarial examples: that of unchanged semantic content. To address this problem, we introduce a more principled notion of an adversarial graph, which is aware of semantic content change. Using Contextual Stochastic Block Models (CSBMs) and real-world graphs, our results suggest: $i)$ for a majority of nodes the prevalent perturbation models include a large fraction of perturbed graphs violating the unchanged semantics assumption; $ii)$ surprisingly, all assessed GNNs show over-robustness - that is robustness beyond the point of semantic change. We find this to be a complementary phenomenon to adversarial examples and show that including the label-structure of the training graph into the inference process of GNNs significantly reduces over-robustness, while having a positive effect on test accuracy and adversarial robustness. Theoretically, leveraging our new semantics-aware notion of robustness, we prove that there is no robustness-accuracy tradeoff for inductively classifying a newly added node.

GNNs suffer from over-robustness, that is robustness beyond the point of semantic change, which can be combated by including the known label-structure at inference time.

Training deep networks on large-scale datasets is computationally challenging. In this work, we explore the problem of ``\textit{how to accelerate adaptive gradient algorithms in a general manner}", and aim to provide practical efficiency-boosting insights. To this end, we propose an effective and general {Weight-decay-Integrated Nesterov acceleration} (Win) to accelerate adaptive algorithms. Taking AdamW and Adam as examples, we minimize a dynamical loss per iteration which combines the vanilla training loss and a dynamic regularizer inspired by proximal point method (PPM) to improve the convexity of the problem. To introduce Nesterov-alike-acceleration into AdamW and Adam, we respectively use the first- and second-order Taylor approximations of vanilla loss to update the variable twice. In this way, we arrive at our Win acceleration for AdamW and Adam that uses a conservative step and a reckless step to update twice and then linearly combines these two updates for acceleration. Next, we extend Win acceleration to LAMB and SGD. Our transparent acceleration derivation could provide insights for other accelerated methods and their integration into adaptive algorithms. Besides, we prove the convergence of Win-accelerated adaptive algorithms and justify their convergence superiority over their non-accelerated counterparts by taking AdamW and Adam as examples. Experimental results testify to the faster convergence speed and superior performance of our Win-accelerated AdamW, Adam, LAMB and SGD over their non-accelerated counterparts on vision classification tasks and language modeling tasks with both CNN and Transformer backbones. We hope Win shall be a default acceleration option for popular optimizers in deep learning community to improve the training efficiency. Code will be released at \url{https://github.com/sail-sg/win}.

We propose a new and general Weight-decay-Integrated Nesterov acceleration for adaptive algorithms to enhance their convergence speed, and also analyze their convergence justify their convergence superiority.

Modern Deep Reinforcement Learning (RL) algorithms require estimates of the maximal Q-value, which are difficult to compute in continuous domains with an infinite number of possible actions. In this work, we introduce a new update rule for online and offline RL which directly models the maximal value using Extreme Value Theory (EVT), drawing inspiration from economics. By doing so, we avoid computing Q-values using out-of-distribution actions which is often a substantial source of error. Our key insight is to introduce an objective that directly estimates the optimal soft-value functions (LogSumExp) in the maximum entropy RL setting without needing to sample from a policy. Using EVT, we derive our \emph{Extreme Q-Learning} framework and consequently online and, for the first time, offline MaxEnt Q-learning algorithms, that do not explicitly require access to a policy or its entropy. Our method obtains consistently strong performance in the D4RL benchmark, outperforming prior works by \emph{10+ points} on the challenging Franka Kitchen tasks while offering moderate improvements over SAC and TD3 on online DM Control tasks. Visualizations and code can be found on our website.

Introduce a novel framework for Q-learning that models the maximal soft-values without needing to sample from a policy and reaches SOTA performance on online and offline RL settings.

Channel squeezing is one of the central operations performed in CNN bottlenecks to reduce the number of channels in a feature map. This operation is carried out by using a 1 × 1 pointwise convolution which constitutes a significant amount of computations and parameters in a given network. ResNet-50 for instance, consists of 16 such layers which form 33% of total layers and 25% (1.05B/4.12B) of total FLOPs or computations. In the light of their predominance, we propose a novel “Fast Adaptive Channel Squeezing” module which carries out the squeezing operation in a computationally efficient manner. The key benefit of FACS is that it neither alters the number of parameters nor affects the accuracy of a given network. When plugged into diverse CNNs architectures, namely ResNet, VGG, and MobileNet-v2, FACS achieves state-of-the-art performance on ImageNet and CIFAR datasets at dramatically reduced FLOPs. FACS also cuts the training time significantly, and lowers the latency which is particularly advantageous for fast inference on edge devices. The source-code will be made publicly available.

Computationally Efficient Channel Squeezing in CNNs with high representation power

The past decade has witnessed a drastic increase in modern deep neural networks (DNNs) size, especially for generative adversarial networks (GANs). Since GANs usually suffer from high computational complexity, researchers have shown an increased interest in applying pruning methods to reduce the training and inference costs of GANs. Among different pruning methods invented for supervised learning, dynamic sparse training (DST) has gained increasing attention recently as it enjoys excellent training efficiency with comparable performance to post-hoc pruning. Hence, applying DST on GANs, where we train a sparse GAN with a fixed parameter count throughout training, seems to be a good candidate for reducing GAN training costs. However, a few challenges, including the degrading training instability, emerge due to the adversarial nature of GANs. Hence, we introduce a quantity called balance ratio (BR) to quantify the balance of the generator and the discriminator. We conduct a series of experiments to show the importance of BR in understanding sparse GAN training. Building upon single dynamic sparse training (SDST), where only the generator is adjusted during training, we propose double dynamic sparse training (DDST) to control the BR during GAN training. Empirically, DDST automatically determines the density of the discriminator and greatly boosts the performance of sparse GANs on multiple datasets.

We propose a quantity named balance ratio to investigate and improve dynamic sparse training for GANs.

Individual privacy accounting enables bounding differential privacy (DP) loss individually for each participant involved in the analysis. This can be informative as often the individual privacy losses are considerably smaller than those indicated by the DP bounds that are based on considering worst-case bounds at each data access. In order to account for the individual losses in a principled manner, we need a privacy accountant for adaptive compositions of mechanisms, where the loss incurred at a given data access is allowed to be smaller than the worst-case loss. This kind of analysis has been carried out for the Rényi differential privacy by Feldman and Zrnic (2021), however not yet for the so-called optimal privacy accountants. We make first steps in this direction by providing a careful analysis using the Gaussian differential privacy which gives optimal bounds for the Gaussian mechanism, one of the most versatile DP mechanisms. This approach is based on determining a certain supermartingale for the hockey-stick divergence and on extending the Rényi divergence-based fully adaptive composition results by Feldman and Zrnic (2021). We also consider measuring the individual $(\varepsilon,\delta)$-privacy losses using the so-called privacy loss distributions. Using the Blackwell theorem, we can then use the results of Feldman and Zrnic (2021) to construct an approximative individual $(\varepsilon,\delta)$-accountant. We also show how to speed up the FFT-based individual DP accounting using the Plancherel theorem.

Accurate privacy analysis of fully adaptive compositions using Gaussian differential privacy

Modern machine learning requires system designers to specify aspects of the learning pipeline, such as losses, architectures, and optimizers. Meta-learning, or learning-to-learn, instead aims to learn those aspects, and promises to unlock greater capabilities with less manual effort. One particularly ambitious goal of meta-learning is to train general purpose learning algorithms from scratch, using only black box models with minimal inductive bias. A general purpose learning algorithm is one which takes in training data, and produces test-set predictions across a wide range of problems, without any explicit definition of an inference model, training loss, or optimization algorithm. In this paper we show that Transformers and other black-box models can be meta-trained to act as general purpose learning algorithms, and can generalize to learn on different datasets than used during meta-training. We characterize phase transitions between algorithms that generalize, algorithms that memorize, and algorithms that fail to meta-train at all, induced by changes in model size, number of tasks used during meta-training, and meta-optimization hyper-parameters. We further show that the capabilities of meta-trained algorithms are bottlenecked by the accessible state size determining the next prediction, unlike standard models which are thought to be bottlenecked by parameter count. Finally, we propose practical interventions such as biasing the training distribution that improve the meta-training and meta-generalization of general purpose learning algorithms.

Transformers and other black-box models can exhibit learning-to-learn that generalizes to significantly different datasets while undergoing multiple phase transitions in terms of their learning behavior.

Data augmentation is used extensively to improve model generalisation. However, reliance on external libraries to implement augmentation methods introduces a vulnerability into the machine learning pipeline. It is well known that backdoors can be inserted into machine learning models through serving a modified dataset to train on. Augmentation therefore presents a perfect opportunity to perform this modification without requiring an initially backdoored dataset. In this paper we present three backdoor attacks that can be covertly inserted into data augmentation. Our attacks each insert a backdoor using a different type of computer vision augmentation transform, covering simple image transforms, GAN-based augmentation, and composition-based augmentation. By inserting the backdoor using these augmentation transforms, we make our backdoors difficult to detect, while still supporting arbitrary backdoor functionality. We evaluate our attacks on a range of computer vision benchmarks and demonstrate that an attacker is able to introduce backdoors through just a malicious augmentation routine.

We present three backdoor attacks that can be covertly inserted into data augmentation functions.

This paper aims to deal with the ignored real-world complexities in prior work on human motion forecasting, emphasizing the social properties of multi-person motion, the diversity of motion and social interactions, and the complexity of articulated motion. To this end, we introduce a novel task of stochastic multi-person 3D motion forecasting. We propose a dual-level generative modeling framework that separately models independent individual motion at the local level and social interactions at the global level. Notably, this dual-level modeling mechanism can be achieved within a shared generative model, through introducing learnable latent codes that represent intents of future motion and switching the codes' modes of operation at different levels. Our framework is general; we instantiate it with different generative models, including generative adversarial networks and diffusion models, and various multi-person forecasting models. Extensive experiments on CMU-Mocap, MuPoTS-3D, and SoMoF benchmarks show that our approach produces diverse and accurate multi-person predictions, significantly outperforming the state of the art.

We introduce a new task of stochastic multi-person 3D motion forecasting, and propose a dual-level generative modeling framework to address this task.

Reconstructing 3D point clouds into triangle meshes is a key problem in computational geometry and surface reconstruction. Point cloud triangulation solves this problem by providing edge information to the input points. Since no vertex interpolation is involved, it is beneficial to preserve sharp details on the surface. Taking advantage of learning-based techniques in triangulation, existing methods enumerate the complete combinations of candidate triangles, which is both complex and inefficient. In this paper, we leverage the duality between a triangle and its circumcenter, and introduce a deep neural network that detects the circumcenters to achieve point cloud triangulation. Specifically, we introduce multiple anchor priors to divide the neighborhood space of each point. The neural network then learns to predict the presences and locations of circumcenters under the guidance of those anchors. We extract the triangles dual to the detected circumcenters to form a primitive mesh, from which an edge-manifold mesh is produced via simple post-processing. Unlike existing learning-based triangulation methods, the proposed method bypasses an exhaustive enumeration of triangle combinations and local surface parameterization. We validate the efficiency, generalization, and robustness of our method on prominent datasets of both watertight and open surfaces. The code and trained models are provided at \url{https://github.com/Ruitao-L/CircNet}.

We present a deep neural architecture that detects circumcenters of triangles in the dual space to reconstruct 3D point clouds into triangular meshes efficiently

Self-supervised pretraining has been extensively studied in language and vision domains, where a unified model can be easily adapted to various downstream tasks by pretraining representations without explicit labels. When it comes to sequential decision-making tasks, however, it is difficult to properly design such a pretraining approach that can cope with both high-dimensional perceptual information and the complexity of sequential control over long interaction horizons. The challenge becomes combinatorially more complex if we want to pretrain representations amenable to a large variety of tasks. To tackle this problem, in this work, we formulate a general pretraining-finetuning pipeline for sequential decision making, under which we propose a generic pretraining framework \textit{Self-supervised Multi-task pretrAining with contRol Transformer (SMART)}. By systematically investigating pretraining regimes, we carefully design a Control Transformer (CT) coupled with a novel control-centric pretraining objective in a self-supervised manner. SMART encourages the representation to capture the common essential information relevant to short-term control and long-term control, which is transferrable across tasks. We show by extensive experiments in DeepMind Control Suite that SMART significantly improves the learning efficiency among seen and unseen downstream tasks and domains under different learning scenarios including Imitation Learning (IL) and Reinforcement Learning (RL). Benefiting from the proposed control-centric objective, SMART is resilient to distribution shift between pretraining and finetuning, and even works well with low-quality pretraining datasets that are randomly collected. The codebase, pretrained models and datasets are provided at https://github.com/microsoft/smart.

We propose a pretraining framework for sequential decision making based on a self-supervised objectives and a control transformer architecture, leading to significantly higher learning efficiency in various downstram control tasks.

Graph-level anomaly detection aims to identify abnormal samples of a set of graphs in an unsupervised manner. It is non-trivial to find a reasonable decision boundary between normal data and anomalous data without using any anomalous data in the training stage, especially for data in graphs. This paper first proposes a novel deep graph-level anomaly detection model, which learns the graph representation with maximum mutual information between substructure features and global structure features while exploring a hypersphere anomaly decision boundary. The deep orthogonal projection layer is adopted to keep the training data distribution consistent with the decision hypersphere thus avoiding erroneous evaluations. We further propose projecting the normal data into the interval region between two co-centered hyperspheres, which makes the normal data distribution more compact and effectively overcomes the issue of outliers falling close to the center of the hypersphere. The numerical and visualization results on a few graph datasets demonstrate the effectiveness and superiority of our methods in comparison to many baselines and state-of-the-art.

A deep orthogonal graph-level anomaly detection method and its improvement.

Generative adversarial networks (GANs) are among the most successful models for learning high-complexity, real-world distributions. However, in theory, due to the highly non-convex, non-concave landscape of the minmax training objective, GAN remains one of the least understood deep learning models. In this work, we formally study how GANs can efficiently learn certain hierarchically generated distributions that are close to the distribution of real-life images. We prove that when a distribution has a structure that we refer to as \emph{forward super-resolution}, then simply training generative adversarial networks using stochastic gradient descent ascent (SGDA) can learn this distribution efficiently, both in sample and time complexities. We also provide empirical evidence that our assumption ``forward super-resolution'' is very natural in practice, and the underlying learning mechanisms that we study in this paper (to allow us efficiently train GAN via GDA in theory) simulates the actual learning process of GANs on real-world problems.

We provide a theory to study how generative adversarial networks (GANs) can efficiently learn certain hierarchically generated distributions that are close to the distribution of images in practice.

Time series modeling is a well-established problem, which often requires that methods (1) expressively represent complicated dependencies, (2) forecast long horizons, and (3) efficiently train over long sequences. State-space models (SSMs) are classical models for time series, and prior works combine SSMs with deep learning layers for efficient sequence modeling. However, we find fundamental limitations with these prior approaches, proving their SSM representations cannot express autoregressive time series processes. We thus introduce SpaceTime, a new state-space time series architecture that improves all three criteria. For expressivity, we propose a new SSM parameterization based on the companion matrix---a canonical representation for discrete-time processes---which enables SpaceTime's SSM layers to learn desirable autoregressive processes. For long horizon forecasting, we introduce a "closed-loop" variation of the companion SSM, which enables SpaceTime to predict many future time-steps by generating its own layer-wise inputs. For efficient training and inference, we introduce an algorithm that reduces the memory and compute of a forward pass with the companion matrix. With sequence length $\ell$ and state-space size $d$, we go from $\tilde{O}(d \ell)$ naïvely to $\tilde{O}(d + \ell)$. In experiments, our contributions lead to state-of-the-art results on extensive and diverse benchmarks, with best or second-best AUROC on 6 / 7 ECG and speech time series classification, and best MSE on 14 / 16 Informer forecasting tasks. Furthermore, we find SpaceTime (1) fits AR($p$) processes that prior deep SSMs fail on, (2) forecasts notably more accurately on longer horizons than prior state-of-the-art, and (3) speeds up training on real-world ETTh1 data by 73% and 80% relative wall-clock time over Transformers and LSTMs.

We propose SpaceTime, a deep state space time series model that achieves state-of-the-art results on forecasting and classification benchmarks, by improving expressiveness, forecasting flexibility, and training efficiency over prior approaches.

Federated learning often suffers from slow and unstable convergence due to heterogeneous characteristics of participating client datasets. Such a tendency is aggravated when the client participation ratio is low since the information collected from the clients is prone to have large variations. To tackle this challenge, we propose a novel federated learning framework, which improves the consistency across clients and facilitates the convergence of the server model. This is achieved by making the server broadcast a global model with a gradient acceleration. By adopting the strategy, the proposed algorithm conveys the projective global update information to participants effectively with no extra communication cost and relieves the clients from storing the previous models. We also regularize local updates by aligning each of the clients with the overshot global model to reduce bias and improve the stability of our algorithm. We perform comprehensive empirical studies on real data under various settings and demonstrate remarkable performance gains of the proposed method in terms of accuracy and communication efficiency compared to the state-of-the-art methods, especially with low client participation rates. We will release our code to facilitate and disseminate our work.

We present, FedACG, a novel federated learning framework which reduce the gap between global and local losses by incorporating the global momentum to guide client updates.

Even pruned by the state-of-the-art network compression methods, recent research shows that deep learning model training still suffers from the demand of massive data usage. In particular, Graph Neural Networks (GNNs) trained upon non-Euclidean graph data often encounter relatively higher time costs, due to its irregular and nasty density properties, compared with data in the regular Euclidean space (e.g., image or text). Another natural property accompanied with graphs is class-imbalance which cannot be alleviated even with massive data, therefore hinders GNNs’ ability in generalization. To fully tackle these unpleasant properties, (i) theoretically, we introduce a hypothesis about to what extent a subset of the training data can approximate the full dataset’s learning effectiveness. The effectiveness is further guaranteed and proved by the gradients’ distance between the subset and the full set; (ii) empirically, we discover that during the learning process of a GNN, some samples in the training dataset are informative in providing gradients to update model parameters. Moreover, the informative subset evolves dynamically during the training process, for samples that are informative in the current training epoch may not be so in the next one. We refer this observation as dynamic data sparsity. We also notice that sparse subnets pruned from a well-trained GNN sometimes forget the information provided by the informative subset, reflected in their poor performance upon the subset. Based on these findings, we develop a unified data-model dynamic sparsity framework named Graph Decantation (GraphDec) to address challenges brought by training upon a massive class-imbalanced graph dataset. The key idea of GraphDec is to identify the informative subset dynamically during the training process by adopting sparse graph contrastive learning. Extensive experiments on multiple benchmark datasets demonstrate that GraphDec outperforms state-of-the-art baselines for the class-imbalanced graph classification and class-imbalanced node classification tasks, with respect to classification accuracy and data usage efficiency.

We propose Graph Decantation, a novel method of discovering unified dynamic sparsity from both GNN model and graph data, to learn balanced graph representations.

Model evaluation is a crucial step in ensuring reliable machine learning systems. Currently, predictive models are evaluated on held-out test data, quantifying aggregate model performance. Limitations of available test data make it challenging to evaluate model performance on small subgroups or when the environment changes. Synthetic test data provides a unique opportunity to address this challenge; instead of evaluating predictive models on real data, we propose to use synthetic data. This brings two advantages. First, supplementing and increasing the amount of evaluation data can lower the variance of model performance estimates compared to evaluation on the original test data. This is especially true for local performance evaluation in low-density regions, e.g. minority or intersectional groups. Second, generative models can be conditioned as to induce a shift in the synthetic data distribution, allowing us to evaluate how supervised models could perform in different target settings. In this work, we propose SYNG4ME: an automated suite of synthetic data generators for model evaluation. By generating smart synthetic data sets, data practitioners have a new tool for exploring how supervised models may perform on subgroups of the data, and how robust methods are to distributional shifts. We show experimentally that SYNG4ME achieves more accurate performance estimates compared to using the test data alone.

Evaluating ML supervised models by generating synthetic test data

Federated learning (FL) is a popular distributed machine learning framework in which clients aggregate models' parameters instead of sharing their individual data. In FL, clients communicate with the server under limited network bandwidth frequently, which arises the communication challenge. To resolve this challenge, multiple compression methods have been proposed to reduce the transmitted parameters. However, these techniques show that the federated performance degrades significantly with Non-IID (non-identically independently distributed) datasets. To address this issue, we propose an effective method, called FedPSE, which solves the efficiency challenge of FL with heterogeneous data. FedPSE compresses the local updates on clients using Top-K sparsification and aggregates these updates on the server by element-wise average. Then clients download the personalized sparse updates from the server to update their individual local models. We then theoretically analyze the convergence of FedPSE under the non-convex setting. Moreover, extensive experiments on four benchmark tasks demonstrate that our FedPSE outperforms the state-of-the-art methods on Non-IID datasets in terms of both efficiency and accuracy.

We propose a federated learning framework to resolve the bidirectional communication challenge on Non-IID datasets.

Generalisation to unseen contexts remains a challenge for embodied navigation agents. In the context of semantic audio-visual navigation (SAVi) tasks, generalisation includes both generalising to unseen indoor visual scenes as well as generalising to unheard sounding objects. Previous SAVi task definitions do not include evaluation conditions on truly novel sounding objects, resorting instead to evaluating agents on unheard sound clips of known objects; meanwhile, previous SAVi methods do not include explicit mechanisms for incorporating domain knowledge about object and region semantics. These weaknesses limit the development and assessment of models' abilities to generalise their learned experience. In this work, we introduce the use of knowledge-driven scene priors in the semantic audio-visual embodied navigation task: we combine semantic information from our novel knowledge graph that encodes object-region relations, spatial knowledge from dual Graph Encoder Networks, and background knowledge from a series of pre-training tasks---all within a reinforcement learning framework for audio-visual navigation. We define a new audio-visual navigation sub-task, where agents are evaluated on novel sounding objects, as opposed to unheard clips of known objects. We show improvements over strong baselines in generalisation to unseen regions and novel sounding objects, within the Habitat-Matterport3D simulation environment, under the SoundSpaces task. We release code, knowledge graphs, and dataset generation details in the supplementary material.

We introduce knowledge-driven scene priors in audio-visual navigation, combining semantics from a knowledge graph, spatial knowledge from dual Graph Encoder Networks, and background knowledge from pre-training tasks—all within an RL framework.

Designing robust loss functions is popular in learning with noisy labels while existing designs did not explicitly consider the overfitting property of deep neural networks (DNNs). As a result, applying these losses may still suffer from overfitting/memorizing noisy labels as training proceeds. In this paper, we first theoretically analyze the memorization effect and show that a lower-capacity model may perform better on noisy datasets. However, it is non-trivial to design a neural network with the best capacity given an arbitrary task. To circumvent this dilemma, instead of changing the model architecture, we decouple DNNs into an encoder followed by a linear classifier and propose to restrict the function space of a DNN by a representation regularizer. Particularly, we require the distance between two self-supervised features to be positively related to the distance between the corresponding two supervised model outputs. Our proposed framework is easily extendable and can incorporate many other robust loss functions to further improve performance. Extensive experiments and theoretical analyses support our claims. Code is available at https://github.com/UCSC-REAL/SelfSup_NoisyLabel.

We theoretically show the memorization effect of DNN with resepct to the model capacity and propose a representation-based regularizer to mitigate the memorization effect.

Recently, Transformer-based methods have achieved surprising performance in the field of long-term series forecasting, but the attention mechanism for computing global correlations entails high complexity. And they do not allow for targeted modeling of local features as CNN structures do. To solve the above problems, we propose to combine local features and global correlations to capture the overall view of time series (e.g., fluctuations, trends). To fully exploit the underlying information in the time series, a multi-scale branch structure is adopted to model different potential patterns separately. Each pattern is extracted with down-sampled convolution and isometric convolution for local features and global correlations, respectively. In addition to being more effective, our proposed method, termed as Multi-scale Isometric Convolution Network (MICN), is more efficient with linear complexity about the sequence length with suitable convolution kernels. Our experiments on six benchmark datasets show that compared with state-of-the-art methods, MICN yields 17.2% and 21.6% relative improvements for multivariate and univariate time series, respectively.

New modeling perspective, new forecasting framework, linear complexity and best performance.

We propose a new method for fine-grained few-shot recognition via deep object parsing. In our framework, an object is made up of $K$ distinct parts and for each part, we learn a dictionary of templates, which is shared across all instances and categories. An object is parsed by estimating the locations of these $K$ parts and a set of active templates that can reconstruct the part features. We recognize test instances by comparing its active templates and the relative geometry of its part locations against those of the presented few-shot instances. Our method is end-to-end trainable to learn part templates on-top of a convolutional backbone. To combat visual distortions such as orientation, pose and size, we learn templates at multiple scales, and at test-time parse and match instances across these scales. We show that our method is competitive with the state-of-the-art, and by virtue of parsing enjoys interpretability as well.

A method for fine-grained few-shot recognition that relies on representing objects as a collection of parts, where each part is identified by a location and a set of active templates.

We address the challenging problem of jointly inferring the 3D flow and volumetric densities moving in a fluid from a monocular input video with a deep neural network. Despite the complexity of this task, we show that it is possible to train the corresponding networks without requiring any 3D ground truth for training. In the absence of ground truth data we can train our model with observations from real-world capture setups instead of relying on synthetic reconstructions. We make this unsupervised training approach possible by first generating an initial prototype volume which is then moved and transported over time without the need for volumetric supervision. Our approach relies purely on image-based losses, an adversarial discriminator network, and regularization. Our method can estimate long-term sequences in a stable manner, while achieving closely matching targets for inputs such as rising smoke plumes.

We train a network to estimate 3D motions and densities from single view videos of smoke without using 3D ground truth.

Latent variable models have become instrumental in computational neuroscience for reasoning about neural computation. This has fostered the development of powerful offline algorithms for extracting latent neural trajectories from neural recordings. However, despite the potential of real-time alternatives to give immediate feedback to experimentalists, and enhance experimental design, they have received markedly less attention. In this work, we introduce the exponential family variational Kalman filter (eVKF), an online recursive Bayesian method aimed at inferring latent trajectories while simultaneously learning the dynamical system generating them. eVKF works for arbitrary likelihoods and utilizes the constant base measure exponential family to model the latent state stochasticity. We derive a closed-form variational analog to the predict step of the Kalman filter which leads to a provably tighter bound on the ELBO compared to another online variational method. We validate our method on synthetic and real-world data, and, notably, show that it achieves competitive performance.

A real-time variational Bayesian method aimed at uncovering latent neural trajectories and their dynamical systems.

To improve the robustness of DRL agents, it is important to study their vulnerability under adversarial attacks that would lead to extreme behaviors desired by adversaries. Preference-based RL (PbRL) aims for learning desired behaviors with human preferences. In this paper, we propose PALM, a preference-based adversarial manipulation method against DRL agents which adopts human preferences to perform targeted attacks with the assistance of an intention policy and a weighting function. The intention policy is trained based on the PbRL framework to guide the adversarial policy to mitigate restrictions of the victim policy during exploration, and the weighting function learns weight assignment to improve the performance of the adversarial policy. Theoretical analysis demonstrates that PALM converges to critical points under some mild conditions. Empirical results on a few manipulation tasks of Meta-world show that PALM exceeds the performance of state-of-the-art adversarial attack methods under the targeted setting. Additionally, we show the vulnerability of the offline RL agents by fooling them into behaving as human desires on several Mujoco tasks. Our code and videos are available in https://sites.google.com/view/palm-adversarial-attack.

A preference-based adversarial attack method that manipulates the victim policy to perform human desired behaviors.

Language models significantly benefit from context tokens, such as prompts or scratchpads. They perform better when prompted with concrete training examples and abstract statements about the target task (instructions), and they acquire new capabilities to perform complex tasks by generating step-by-step reasoning (scratch-pad) before predicting the final answers. However, they do not internalize these performance gains, which disappear when the context tokens are gone. Consequently, we always need to pay extra computation for this gain, and it is unclear how to transfer the capabilities acquired by context tokens to other tasks, or how to leverage the context tokens when their length exceeds the context window size. Our work proposes to apply context distillation so that a language model can internalize these gains. Concretely, given an input for the target task, we let the model use all relevant context tokens to generate the output, using ``[instructions] + [task-input]'' to predict ``[scratch-pad] + [final answer]''; then we fine-tune the same model to predict the above ``[final answer]'' conditioned on the ``[task-input]'', without seeing the ``[instructions]'' or using the ``[scratch-pad]''. This incentivizes the model to behave as if the context were present, hence updating the parameters to internalize the context information. We show that context distillation can be used as a general method for learning. In particular, we demonstrate that context distillation can effectively internalize 3 types of contexts: 1) abstract task instructions and natural language explanations of why an output is correct or incorrect on Natural-Instructions-V2; 2) step-by-step reasoning on 8-digit addition questions, where we show the model can apply this newly acquired capability to downstream question answering tasks; and 3) concrete training examples on the SPIDER Text-to-SQL dataset, where context distillation outperforms directly learning with gradient descent by 7%.

applying context distillation as a general tool to translate language into parameter updates; used to distill instructions, explanations, examples, knowledge, and reasoning procedure.

The recently-introduced Lottery Ticket Hypothesis (LTH) posits that it is possible to extract a sparse trainable subnetwork from a dense network using iterative magnitude pruning. By iteratively training the model, removing the connections with the lowest global weight magnitude and rewinding the remaining connections, sparse networks can be extracted that, when fully trained, reach a similar or better performance than their dense counterpart. Intuitively, this approach of comparing connection weights globally removes a lot of context about the connection weights and their relations to other connections in their layer as the weight distributions in layers throughout the network often differ significantly. In this paper we study a number of different approaches that try to recover some of this layer distributional context by computing an importance value for each connection that is dependent on the weights of the other connections in the same layer. We then generalise the LTH to use weight importances rather than weight magnitudes. Experiments using these importance metrics on several architectures and datasets, reveal interesting aspects on the structure and emergence of Lottery tickets. We find that given a repeatable training procedure, applying different importance metrics lead to distinct performant lottery tickets with little overlapping connections.

Using different importance measures in the LTH procedure to determine properties of the resulting LTs.

Witnessing the impressive achievements of pre-training techniques on large-scale data in the field of computer vision and natural language processing, we wonder whether this idea could be adapted in a grab-and-go spirit, and mitigate the sample inefficiency problem for visuomotor driving. Given the highly dynamic and variant nature of the input, the visuomotor driving task inherently lacks the view and translation invariance, and the visual input contains massive irrelevant information for decision making, resulting in predominant pre-training approaches from general vision less suitable for the autonomous driving task. To this end, we propose PPGeo (Policy Pre-training via Geometric modeling), an intuitive and straightforward fully self-supervised framework curated for the policy pre-training in visuomotor driving. We aim at learning policy representations as a powerful abstraction by modeling 3D geometric scenes on large-scale unlabeled and uncalibrated YouTube driving videos. The proposed PPGeo is performed in two stages to support effective self-supervised training. In the first stage, the geometric modeling framework generates pose and depth predictions simultaneously, with two consecutive frames as input. In the second stage, the visual encoder learns driving policy representation by predicting the future ego-motion and optimizing with the photometric error based on current visual observation only. As such, the pre-trained visual encoder is equipped with rich driving policy related representations and thereby competent for multiple visuomotor driving tasks. As a side product, the pre-trained geometric modeling networks could bring further improvement to the depth and odometry estimation tasks. Extensive experiments covering a wide span of challenging scenarios have demonstrated the superiority of our proposed approach, where improvements range from 2% to even over 100% with very limited data.

We introduce a visuomotor driving policy pre-training paradigm, which leverages self-supervised geometric modeling to learn driving policy representation and achieves superior performance on various downstream driving tasks.

Graph Neural Networks (GNNs) are local aggregators that derive their expressive power from their sensitivity to network structure. However, this sensitivity comes at a cost: noisy edges degrade performance. In response, many GNNs include edge-weighting mechanisms that scale the contribution of each edge in the aggregation step. However, to account for neighborhoods of varying size, node-embedding mechanisms must normalize these edge-weights across each neighborhood. As such, the impact of noisy edges cannot be eliminated without removing those edges altogether. Motivated by this issue, we introduce the Graph Learning Attention Mechanism (GLAM): a drop-in, differentiable structure learning layer for GNNs that separates the distinct tasks of structure learning and node embedding. In contrast to existing graph learning approaches, GLAM does not require the addition of exogenous structural regularizers or edge-selection heuristics to learn optimal graph structures. In experiments on citation and co-purchase datasets, we demonstrate that our approach can match state of the art semi-supervised node classification accuracies while inducing an order of magnitude greater sparsity than existing graph learning methods.

We introduce a drop-in, differentiable graph structure learning layer for use with GNNs.

Recently, Graph Neural Networks (GNNs) have significantly advanced the performance of machine learning tasks on graphs. However, this technological breakthrough makes people wonder: how does a GNN make such decisions, and can we trust its prediction with high confidence? When it comes to some critical fields, such as biomedicine, where making wrong decisions can have severe consequences, it is crucial to interpret the inner working mechanisms of GNNs before applying them. In this paper, we propose a model-agnostic model-level explanation method for different GNNs that follow the message passing scheme, GNNInterpreter, to explain the high-level decision-making process of the GNN model. More specifically, GNNInterpreter learns a probabilistic generative graph distribution that produces the most discriminative graph pattern the GNN tries to detect when making a certain prediction by optimizing a novel objective function specifically designed for the model-level explanation for GNNs. Compared to existing works, GNNInterpreter is more flexible and computationally efficient in generating explanation graphs with different types of node and edge features, without introducing another blackbox or requiring manually specified domain-specific rules. In addition, the experimental studies conducted on four different datasets demonstrate that the explanation graphs generated by GNNInterpreter match the desired graph pattern if the model is ideal; otherwise, potential model pitfalls can be revealed by the explanation.

We propose a model-level explanation method for GNNs, which is more general, flexible, and computationally efficient than the current SOTA.

We propose a theoretical framework to analyze semi-supervised classification under the low density separation assumption in a high-dimensional regime. In particular, we introduce QLDS, a linear classification model, where the low density separation assumption is implemented via quadratic margin maximization. The algorithm has an explicit solution with rich theoretical properties, and we show that particular cases of our algorithm are the least-square support vector machine in the supervised case, the spectral clustering in the fully unsupervised regime, and a class of semi-supervised graph-based approaches. As such, QLDS establishes a smooth bridge between these supervised and unsupervised learning methods. Using recent advances in the random matrix theory, we formally derive a theoretical evaluation of the classification error in the asymptotic regime. As an application, we derive a hyperparameter selection policy that finds the best balance between the supervised and the unsupervised terms of our learning criterion. Finally, we provide extensive illustrations of our framework, as well as an experimental study on several benchmarks to demonstrate that QLDS, while being computationally more efficient, improves over cross-validation for hyperparameter selection, indicating a high promise of the usage of random matrix theory for semi-supervised model selection.

We introduce a semi-supervised learning algorithm based on the low density assumption and propose a theoretical analysis of the latter.

Diffusion models (DMs) have recently emerged as a promising method in image synthesis. They have surpassed generative adversarial networks (GANs) in both diversity and quality, and have achieved impressive results in text-to-image modeling. However, to date, only little attention has been paid to the detection of DM-generated images, which is critical to prevent adverse impacts on our society. While prior works have shown that GAN-generated images can be reliably detected using automated methods, it is unclear whether the same methods are effective against DMs. In this work, we address this challenge and take a first look at detecting DM-generated images. We approach the problem from two different angles: First, we evaluate the performance of state-of-the-art detectors on a variety of DMs. Second, we analyze DM-generated images in the frequency domain and study different factors that influence the spectral properties of these images. Most importantly, we demonstrate that GANs and DMs produce images with different characteristics, which requires adaptation of existing classifiers to ensure reliable detection. We believe this work provides the foundation and starting point for further research to detect DM deepfakes effectively.

We take a first look at the detection of images generated by diffusion models by evaluating state-of-the-art detectors and analyzing DM-generated images in the frequency domain.

Large language models have recently shown promising progress in mathematical reasoning when fine-tuned with human-generated sequences walking through a sequence of solution steps. However, the solution sequences are not formally structured and the resulting model-generated sequences may not reflect the kind of systematic reasoning we might expect an expert human to produce. In this paper, we study how to build stronger reasoning capability in language models using the idea of relational abstractions. We introduce new types of sequences that more explicitly provide an abstract characterization of the transitions through intermediate solution steps to the goal state. We found that models that are supplied with such sequences as prompts can solve tasks with a significantly higher accuracy, and models that are trained to produce such sequences solve problems better than those that are trained with previously used human-generated sequences and other baselines. Our work thus takes several steps toward elucidating and improving how language models perform on tasks requiring multi-step mathematical reasoning.

Sequences with abstract relations can help models solve mathematical reasoning tasks with a significantly higher accuracy compared to those that are trained with human-generated sequences and other baselines.

We present a learning method that produces a mapping between arbitrary densities, such that random samples of a density can be mapped to random samples of another. In practice, our method is similar to deterministic diffusion processes where samples of the target density are blended with Gaussian noise. The originality of our approach is that, in contrast to several recent works, we do not rely on Langevin dynamics or score-matching concepts. We propose a simpler take on the topic, which is based solely on Bayes' theorem. By studying blended samples and their posteriors, we show that iteratively blending and deblending samples produces random paths between arbitrary densities. We prove that, for finite-variance densities, these paths converge towards a deterministic mapping that can be learnt with a neural network trained to deblend samples. Our method can thus be seen as a generalization of deterministic denoising diffusion where, instead of learning to denoise Gaussian noise, we learn to deblend arbitrary data.

Deriving a deterministic denoising diffusion with very basic concepts (no langevin equation, no score, etc.)

Tail averaging improves on Polyak averaging's non-asymptotic behaviour by excluding a number of leading iterates of stochastic optimization from its calculations. In practice, with a finite number of optimization steps and a learning rate that cannot be annealed to zero, tail averaging can get much closer to a local minimum point of the training loss than either the individual iterates or the Polyak average. However, the number of leading iterates to ignore is an important hyperparameter, and starting averaging too early or too late leads to inefficient use of resources or suboptimal solutions. Setting this hyperparameter to improve generalization is even more difficult, especially in the presence of other hyperparameters and overfitting. Furthermore, before averaging starts, the loss is only weakly informative of the final performance, which makes early stopping unreliable. To alleviate these problems, we propose an anytime variant of tail averaging, that has no hyperparameters and approximates the optimal tail at all optimization steps. Our algorithm is based on two running averages with adaptive lengths bounded in terms of the optimal tail length, one of which achieves approximate optimality with some regularity. Requiring only the additional storage for two sets of weights and periodic evaluation of the loss, the proposed two-tailed averaging algorithm is a practical and widely applicable method for improving stochastic optimization.

New approximately optimal iterate averaging algorithm with no hyperparameters that approximates the optimal average at all optimization steps.

This paper investigates a new approach to model-based reinforcement learning using background planning: mixing (approximate) dynamic programming updates and model-free updates, similar to the Dyna architecture. Background planning with learned models is often worse than model-free alternatives, such as Double DQN, even though the former uses significantly more memory and computation. The fundamental problem is that learned models can be inaccurate and often generate invalid states, especially when iterated many steps. In this paper, we avoid this limitation by constraining background planning to a set of (abstract) subgoals and learning only local, subgoal-conditioned models. This goal-space planning (GSP) approach is more computationally efficient, naturally incorporates temporal abstraction for faster long-horizon planning and avoids learning the transition dynamics entirely. We show that our GSP algorithm can learn significantly faster than a Double DQN baseline in a variety of situations.

A new approach to model-based RL where planning operates in an abstract subgoal space

Deep neural networks (DNN) has shown great capacity of modeling a dynamical system, but these DNN-based dynamical models usually do not obey conservation laws. To impose the learned DNN dynamical models with key physical properties such as conservation laws, this paper proposes a two-step approach to endow the invariant priors into the simulations. We first establish a contrastive learning framework to capture the system invariants along the trajectory observations. During the dynamics modeling, we design a projection layer of DNNs to preserve the system invariance. Through experiments, we show our method consistently outperforms the baseline in both coordinate error and conservation metrics and can be further extended to complex and large dynamics by leveraging autoencoder. Notably, a byproduct of our framework is the automated conservation law discovery for dynamical systems with single conservation property.

We learn dynamical system conservation property through contrastive learning and impose it during simulation to improve prediction robustness.

Feature selection is the problem of selecting a subset of features for a machine learning model that maximizes model quality subject to a budget constraint. For neural networks, prior methods, including those based on $\ell_1$ regularization, attention, and other techniques, typically select the entire feature subset in one evaluation round, ignoring the residual value of features during selection, i.e., the marginal contribution of a feature given that other features have already been selected. We propose a feature selection algorithm called Sequential Attention that achieves state-of-the-art empirical results for neural networks. This algorithm is based on an efficient one-pass implementation of greedy forward selection and uses attention weights at each step as a proxy for feature importance. We give theoretical insights into our algorithm for linear regression by showing that an adaptation to this setting is equivalent to the classical Orthogonal Matching Pursuit (OMP) algorithm, and thus inherits all of its provable guarantees. Our theoretical and empirical analyses offer new explanations towards the effectiveness of attention and its connections to overparameterization, which may be of independent interest.

Sequential feature selection using the attention mechanism, with provable guarantees.

Decision makers often wish to use offline historical data to compare sequential-action policies at various world states. Importantly, computational tools should produce confidence values for such offline policy comparison (OPC) to account for statistical variance and limited data coverage. Nevertheless, there is little work that directly evaluates the quality of confidence values for OPC. In this work, we address this issue by creating benchmarks for OPC with Confidence (OPCC), derived by adding sets of policy comparison queries to datasets from offline reinforcement learning. In addition, we present an empirical evaluation of the "risk versus coverage" trade-off for a class of model-based baselines. In particular, the baselines learn ensembles of dynamics models, which are used in various ways to produce simulations for answering queries with confidence values. While our results suggest advantages for certain baseline variations, there appears to be significant room for improvement in future work.

We introduce a benchmark and baselines to study uncertainty estimation via policy comparisons in offline reinforcement learning datasets.

Graph neural networks (GNNs) learn node representations by recursively aggregating neighborhood information on graph data. However, in the federated setting, data samples (nodes) located in different clients may be connected to each other, leading to huge information loss to the training method. Existing federated graph learning frameworks solve such a problem by generating missing neighbors or sending information across clients directly. None are suitable for training deep GNNs, which require a more expansive receptive field and higher communication costs. In this work, we introduce a novel framework named $Fed^2GNN$ for federated graph learning of deep GNNs via reconstructing neighborhood information of nodes. Specifically, we design a graph structure named rooted tree. The node embedding obtained by encoding on the rooted tree is the same as that obtained by encoding on the induced subgraph surrounding the node, which allows us to reconstruct the neighborhood information by building the rooted tree of the node. An encoder-decoder framework is then proposed, wherein we first encode missing neighbor information and then decode it to build the rooted tree. Extensive experiments on real-world network datasets show the effectiveness of our framework for training deep GNNs while also achieving better performance for training shadow GNN models

We study the problem of graph representation learning under a federated setting and propose a novel framework for federated learning of deep graph neural networks via reconstructing neighborhood information of nodes.

In the traditional federated learning setting, a central server coordinates a network of clients to train one global model. However, the global model may serve many clients poorly due to data heterogeneity. Moreover, there may not exist a trusted central party that can coordinate the clients to ensure that each of them can benefit from others. To address these concerns, we present a novel decentralized framework, FedeRiCo, where each client can learn as much or as little from other clients as is optimal for its local data distribution. Based on expectation-maximization, FedeRiCo estimates the utilities of other participants’ models on each client’s data so that everyone can select the right collaborators for learning. As a result, our algorithm outperforms other federated, personalized, and/or decentralized approaches on several benchmark datasets, being the only approach that consistently performs better than training with local data only.

We propose a novel personalized decentralized federated learning framework for heterogeneous client data by collaborating with the right clients.

This paper studies a curious phenomenon that machine learning model with Transformer architectures have sparse activation maps. By activation map we refer to the intermediate output of the multi-layer perceptrons (MLPs) after a ReLU activation function, and by "sparse" we mean that on average very few entries (e.g., 3.0% for T5-Base and 6.3% for ViT-B16) are nonzero for each input to MLP. Moreover, larger Transformers with more layers and wider MLP hidden dimensions are sparser as measured by the percentage of nonzero entries. Through extensive experiments we demonstrate that the emergence of sparsity is a prevalent phenomenon that occurs for both natural language processing and vision tasks, on both training and evaluation data, for Transformers of various configurations, at layers of all depth levels. We discuss how sparsity immediately implies a way to significantly reduce the FLOP count and improve efficiency for Transformers. Moreover, we demonstrate perhaps surprisingly that enforcing an even sparser activation via Top-k thresholding with a small k brings a collection of desired properties, namely less sensitivity to noisy training data, more robustness to input corruptions, and better calibration for their prediction confidence.

Learned Transformers for NLP (e.g., T5) and Vision (e.g., ViT) tasks produce sparse representations in their MLP layers. The sparsity may be leveraged to improve robustness, calibration, and computational efficiency of Transformer models.

Class imbalance problems frequently occur in real-world tasks, and conventional deep learning algorithms are well known for performance degradation on imbalanced training datasets. To mitigate this problem, many approaches have aimed to balance among given classes by re-weighting or re-sampling training samples. These re-balancing methods increase the impact of minority classes and reduce the influence of majority classes on the output of models. However, the extracted representations may be of poor quality owing to the limited number of minority samples. To handle this restriction, several methods have been developed that increase the representations of minority samples by leveraging the features of the majority samples. Despite extensive recent studies, no deep analysis has been conducted on determination of classes to be augmented and strength of augmentation has been conducted. In this study, we first investigate the correlation between the degree of augmentation and class-wise performance, and find that the proper degree of augmentation must be allocated for each class to mitigate class imbalance problems. Motivated by this finding, we propose a simple and efficient novel curriculum, which is designed to find the appropriate per-class strength of data augmentation, called CUDA: CUrriculum of Data Augmentation for long-tailed recognition. CUDA can simply be integrated into existing long-tailed recognition methods. We present the results of experiments showing that CUDA effectively achieves better generalization performance compared to the state-of-the-art method on various imbalanced datasets such as CIFAR-100-LT, ImageNet-LT, and iNaturalist 2018.

We propose a class-wise data augmentation method by designing the curriculum of data augmentation, which is based on our findings that stronger augmentation on major classes improves the performance on long-tailed recognition.

Intelligent systems deployed in the real world suffer from catastrophic forgetting when exposed to a sequence of tasks. Humans, on the other hand, acquire, consolidate, and transfer knowledge between tasks that rarely interfere with the consolidated knowledge. Accompanied by self-regulated neurogenesis, continual learning in the brain is governed by the rich set of neurophysiological processes that harbor different types of knowledge which are then integrated by the conscious processing. Thus, inspired by Global Workspace Theory of conscious information access in the brain, we propose TAMiL, a continual learning method that entails task-attention modules to capture task-specific information from the common representation space. We employ simple, undercomplete autoencoders to create a communication bottleneck between the common representation space and the global workspace, allowing only the task-relevant information to the global workspace, thereby greatly reducing task interference. Experimental results show that our method outperforms state-of-the-art rehearsal-based and dynamic sparse approaches and bridges the gap between fixed capacity and parameter isolation approaches while being scalable. We also show that our method effectively mitigates catastrophic forgetting while being well-calibrated with reduced task-recency bias.

A continual learning method that entails task-attention modules to capture task-specific information from the common representation space

Self-Supervised Learning (SSL) methods such as VICReg, Barlow Twins or W-MSE avoid collapse of their joint embedding architectures by constraining or regularizing the covariance matrix of their projector’s output. This study highlights important properties of such strategy, which we coin Variance-Covariance regularization (VCReg). More precisely, we show that VCReg enforces pairwise independence between the features of the learned representation. This result emerges by bridging VCReg applied on the projector’s output to kernel independence criteria applied on the projector’s input. This provides the first theoretical motivations and explanations of VCReg. We empirically validate our findings where (i) we put in evidence which projector’s characteristics favor pairwise independence, (ii) we use these findings to obtain nontrivial performance gains for VICReg, (iii) we demonstrate that the scope of VCReg goes beyond SSL by using it to solve Independent Component Analysis. We hope that our findings will support the adoption of VCReg in SSL and beyond.

We study how SSL methods such as VICReg and Barlow Twins enforce pairwise independence of representations via their Variance Covariance regularization (VCReg), improve VICReg using our findings and show VCReg to be beneficial outside of SSL.

Empirical risk minimization (ERM) is known to be non-robust in practice to distributional shift where the training and the test distributions are different. A suite of approaches, such as importance weighting, and variants of distributionally robust optimization (DRO), have been proposed to solve this problem. But a line of recent work has empirically shown that these approaches do not significantly improve over ERM in real applications with distribution shift. The goal of this work is to obtain a comprehensive theoretical understanding of this intriguing phenomenon. We first posit the class of Generalized Reweighting (GRW) algorithms, as a broad category of approaches that iteratively update model parameters based on iterative reweighting of the training samples. We show that when overparameterized models are trained under GRW, the resulting models are close to that obtained by ERM. We also show that adding small regularization which does not greatly affect the empirical training accuracy does not help. Together, our results show that a broad category of what we term GRW approaches are not able to achieve distributionally robust generalization. Our work thus has the following sobering takeaway: to make progress towards distributionally robust generalization, we either have to develop non-GRW approaches, or perhaps devise novel classification/regression loss functions that are adapted to GRW approaches.

We theoretically prove that the broad family of GRW algorithms which covers many popular methods has an implicit bias almost equivalent to ERM, and thus cannot improve over ERM.

All state-of-the-art (SOTA) differentially private machine learning (DP ML) methods are iterative in nature, and their privacy analyses allow publicly releasing the intermediate training checkpoints. However, DP ML benchmarks, and even practical deployments, typically use only the final training checkpoint to make predictions. In this work, for the first time, we comprehensively explore various methods that aggregate intermediate checkpoints to improve the utility of DP training. Empirically, we demonstrate that checkpoint aggregations provide significant gains in the prediction accuracy over the existing SOTA for CIFAR10 and StackOverflow datasets, and that these gains get magnified in settings with periodically varying training data distributions. For instance, we improve SOTA StackOverflow accuracies to 22.7\% (+0.43\% absolute) for $\epsilon=8.2$, and 23.84\% (+0.43\%) for $\epsilon=18.9$. Theoretically, we show that uniform tail averaging of checkpoints improves the empirical risk minimization bound compared to the last checkpoint of DP-SGD. Lastly, we initiate an exploration into estimating the uncertainty that DP noise adds in the predictions of DP ML models. We prove that, under standard assumptions on the loss function, the sample variance from last few checkpoints provides a good approximation of the variance of the final model of a DP run. Empirically, we show that the last few checkpoints can provide a reasonable lower bound for the variance of a converged DP model.

DP-ML benchmarks and deployments typically use only the final model to make predictions. In this work, for the first time, we comprehensively explore various methods that aggregate intermediate checkpoints to improve the utility of DP training.

In this paper we study online Reinforcement Learning (RL) in partially observable dynamical systems. We focus on the Predictive State Representations (PSRs) model, which is an expressive model that captures other well-known models such as Partially Observable Markov Decision Processes (POMDP). PSR represents the states using a set of predictions of future observations and is defined entirely using observable quantities. We develop a novel model-based algorithm for PSRs that can learn a near optimal policy in sample complexity scaling polynomially with respect to all the relevant parameters of the systems. Our algorithm naturally works with function approximation to extend to systems with potentially large state and observation spaces. We show that given a realizable model class, the sample complexity of learning the near optimal policy only scales polynomially with respect to the statistical complexity of the model class, without any explicit polynomial dependence on the size of the state and observation spaces. Notably, our work is the first work that shows polynomial sample complexities to compete with the globally optimal policy in PSRs. Finally, we demonstrate how our general theorem can be directly used to derive sample complexity bounds for special models including $m$-step weakly revealing and $m$-step decodable tabular POMDPs, POMDPs with low-rank latent transition, and POMDPs with linear emission and latent transition.

PAC learning for PSRs.

Differentiable rendering allows the application of computer graphics on vision tasks, e.g. object pose and shape fitting, via analysis-by-synthesis, where gradients at occluded regions are important when inverting the rendering process.To obtain those gradients, state-of-the-art (SoTA) differentiable renderers use rasterization to collect a set of nearest components for each pixel and aggregate them based on the viewing distance. In this paper, we propose VoGE, which uses ray tracing to capture nearest components with their volume density distributions on the rays and aggregates via integral of the volume densities based on Gaussian ellipsoids, which brings more efficient and stable gradients. To efficiently render via VoGE, we propose an approximate close-form solution for the volume density aggregation and a coarse-to-fine rendering strategy. Finally, we provide a CUDA implementation of VoGE, which gives a competitive rendering speed in comparison to PyTorch3D. Quantitative and qualitative experiment results show VoGE outperforms SoTA counterparts when applied to various vision tasks, e.g., object pose estimation, shape/texture fitting, and occlusion reasoning. The VoGE code is available at: https://github.com/Angtian/VoGE.

VoGE is a differentiable renderer based on ray tracing volume densities, which gives better gradients for occlusion reasoning and yields better pose estimation results.

As safety violations can lead to severe consequences in real-world applications, the increasing deployment of Reinforcement Learning (RL) in safety-critical domains such as robotics has propelled the study of safe exploration for reinforcement learning (safe RL). In this work, we propose a risk preventive training method for safe RL, which learns a statistical contrastive classifier to predict the probability of a state-action pair leading to unsafe states. Based on the predicted risk probabilities, we can collect risk preventive trajectories and reshape the reward function with risk penalties to induce safe RL policies. We conduct experiments in robotic simulation environments. The results show the proposed approach has comparable performance with the state-of-the-art model-based methods and outperforms conventional model-free safe RL approaches.

We propose a contrastive risk prediction method to train safe RL agents with risk preventive trajectory exploration and reward shaping.

Having reliable specifications is an unavoidable challenge in achieving verifiable correctness, robustness, and interpretability of AI systems. Existing specifications for neural networks are in the flavor of “data as specification”, that is, the local neighborhood centering around a reference input is considered to be correct (or robust). However, our empirical study shows that such specifications fail to certify any test data points, making it impractical for real-world applications. We propose a new family of specifications called “neural representation as specification”, which uses the intrinsic information of neural networks — neural activation patterns (NAP) rather than input data to specify the correctness and/or robustness of neural network predictions. We present a simple statistical approach to extracting dominant neural activation patterns. We analyze NAPs from a statistical point of view and find that a single NAP can cover a large number of training and testing data points whereas ad hoc data-as-specification can only cover a single training data point and often zero testing data points. To show the effectiveness of discovered NAPs, we formally verify several important properties, such as a particular type of misclassification never happens for a given NAP, and there is no ambiguity among different NAPs. We show that by using NAP, we can verify the prediction of the entire input space, while still recalling 84% of the data. Thus, we argue that using NAPs is a more reliable and extensible specification for neural network verification.

We propose a new family of specifications based on neural activation patterns and evaluate its effectiveness through both statistical analysis and formal verification.

Deep reinforcement learning agents are notoriously sample inefficient, which considerably limits their application to real-world problems. Recently, many model-based methods have been designed to address this issue, with learning in the imagination of a world model being one of the most prominent approaches. However, while virtually unlimited interaction with a simulated environment sounds appealing, the world model has to be accurate over extended periods of time. Motivated by the success of Transformers in sequence modeling tasks, we introduce IRIS, a data-efficient agent that learns in a world model composed of a discrete autoencoder and an autoregressive Transformer. With the equivalent of only two hours of gameplay in the Atari 100k benchmark, IRIS achieves a mean human normalized score of 1.046, and outperforms humans on 10 out of 26 games, setting a new state of the art for methods without lookahead search. To foster future research on Transformers and world models for sample-efficient reinforcement learning, we release our code and models at https://github.com/eloialonso/iris.

We introduce a data-efficient agent that learns in a world model composed of a discrete autoencoder and an autoregressive Transformer.

The remarkable successes of neural networks in a huge variety of inverse problems have fueled their adoption in disciplines ranging from medical imaging to seismic analysis over the past decade. However, the high dimensionality of such inverse problems has simultaneously left current theory, which predicts that networks should scale exponentially in the dimension of the problem, unable to explain why the seemingly small networks used in these settings work as well as they do in practice. To reduce this gap between theory and practice, a general method for bounding the complexity required for a neural network to approximate a Lipschitz function on a high-dimensional set with a low-complexity structure is provided herein. The approach is based on the observation that the existence of a linear Johnson-Lindenstrauss embedding $\mathbf{A} \in \mathbb{R}^{d \times D}$ of a given high-dimensional set $\mathcal{S} \subset \mathbb{R}^D$ into a low dimensional cube $[-M,M]^d$ implies that for any Lipschitz function $f : \mathcal{S} \to \mathbb{R}^p$, there exists a Lipschitz function $g : [-M,M]^d \to \mathbb{R}^p$ such that $g(\mathbf{A}\mathbf{x}) = f(\mathbf{x})$ for all $\mathbf{x} \in \mathcal{S}$. Hence, if one has a neural network which approximates $g : [-M,M]^d \to \mathbb{R}^p$, then a layer can be added which implements the JL embedding $\mathbf{A}$ to obtain a neural network which approximates $f : \mathcal{S} \to \mathbb{R}^p$. By pairing JL embedding results along with results on approximation of Lipschitz functions by neural networks, one then obtains results which bound the complexity required for a neural network to approximate Lipschitz functions on high dimensional sets. The end result is a general theoretical framework which can then be used to better explain the observed empirical successes of smaller networks in a wider variety of inverse problems than current theory allows.

We provide neural network approximation guarantees for Lipschitz functions on low-complexity sets in high dimensions.

Fairness researchers in machine learning (ML) have coalesced around several fairness criteria which provide formal definitions of what it means for an ML model to be fair. However, these criteria have some serious limitations. We identify four key shortcomings of these formal fairness criteria and address them by extending performative prediction to include a distributionally robust objective. Performative prediction is a recent framework developed to understand the effects of when deploying model influences the distribution on which it is making predictions. We prove a convergence result for our proposed repeated distributionally robust optimization (RDRO). We further verify our results empirically and develop experiments to demonstrate the impact of using RDRO on learning fair ML models.

We develop a model for long-term fairness by considering a distributionally robust optimization objective in the performative prediction framework.

In real-world applications, deep learning models often run in non-stationary environments where the target data distribution continually shifts over time. There have been numerous domain adaptation (DA) methods in both online and offline modes to improve cross-domain adaptation ability. However, these DA methods typically only provide good performance after a long period of adaptation, and perform poorly on new domains before and during adaptation – in what we call the “Unfamiliar Period”, especially when domain shifts happen suddenly and significantly. On the other hand, domain generalization (DG) methods have been proposed to improve the model generalization ability on unadapted domains. However, existing DG works are ineffective for continually changing domains due to severe catastrophic forgetting of learned knowledge. To overcome these limitations of DA and DG in handling the Unfamiliar Period during continual domain shift, we propose RaTP, a framework that focuses on improving models’ target domain generalization (TDG) capability, while also achieving effective target domain adaptation (TDA) capability right after training on certain domains and forgetting alleviation (FA) capability on past domains. RaTP includes a training-free data augmentation module to prepare data for TDG, a novel pseudo-labeling mechanism to provide reliable supervision for TDA, and a prototype contrastive alignment algorithm to align different domains for achieving TDG, TDA and FA. Extensive experiments on Digits, PACS, and DomainNet demonstrate that RaTP significantly outperforms state-of-the-art works from Continual DA, Source-Free DA, Test-Time/Online DA, Single DG, Multiple DG and Unified DA&DG in TDG, and achieves comparable TDA and FA capabilities.

To tackle continual domain shift in real-world applications, this work proposes a novel framework for achieving target domain generalization, target domain adaptation, and forgetting compensation at the same time.

Zeroth-order methods are useful in solving black-box optimization and reinforcement learning problems in unknown environments. It uses function values to estimate the gradient. As optimization problems are often nonconvex, it is a natural question to understand how zeroth-order methods escape saddle points. In this paper, we consider zeroth-order methods, that at each iteration, may freely choose 2$m$ function evaluations where $m$ ranges from 1 to $d$, with $d$ denoting the problem dimension. We show that by adding an appropriate isotropic perturbation at each iteration, a zeroth-order algorithm based on $2m$ function evaluations per iteration can not only find $\epsilon$-second order stationary points polynomially fast, but do so using only $\tilde{O}(\frac{d}{\epsilon^{2.5}})$ function evaluations.

We provide the first result showing that zeroth-order optimization with constant number of function evaluations per iteration can escape saddle points efficiently.

Leading graph contrastive learning (GCL) frameworks conform to the invariance mechanism by encouraging insensitivity to different augmented views of the same graph. Despite the promising performance, invariance worsens representation when augmentations cause aggressive semantics shifts. For example, dropping the super-node can dramatically change a social network's topology. In this case, encouraging invariance to the original graph can bring together dissimilar patterns and hurt the task of instance discrimination. To resolve the problem, we get inspiration from equivariant self-supervised learning and propose Equivariant Graph Contrastive Learning (E-GCL) to encourage the sensitivity to global semantic shifts. Viewing each graph as a transformation to others, we ground the equivariance principle as a cross-graph augmentation -- graph interpolation -- to simulate global semantic shifts. Without using annotation, we supervise the representation of cross-graph augmented views by linearly combining the representations of their original samples. This simple but effective equivariance principle empowers E-GCL with the ability of cross-graph discrimination. It shows significant improvements over the state-of-the-art GCL models in unsupervised learning and transfer learning. Further experiments demonstrate E-GCL's generalization to various graph pre-training frameworks. Code is available at \url{https://anonymous.4open.science/r/E-GCL/}

We propose a cross-graph augmentation to achieve equivariant self-supervised learning on graphs.

Unsupervised meta-learning (UML) essentially shares the spirit of self-supervised learning (SSL) in that their goal aims at learning models without any human supervision so that the models can be adapted to downstream tasks. Further, the learning objective of self-supervised learning, which pulls positive pairs closer and repels negative pairs, also resembles metric-based meta-learning. Metric-based meta-learning is one of the most successful meta-learning methods, which learns to minimize the distance between representations from the same class. One notable aspect of metric-based meta-learning, however, is that it is widely interpreted as a set-level problem since the inference of discriminative class prototypes (or set representations) from few examples is crucial for the performance of downstream tasks. Motivated by this, we propose Set-SimCLR, a novel self-supervised set representation learning framework for targeting UML problem. Specifically, our Set-SimCLR learns a set encoder on top of instance representations to maximize the agreement between two sets of augmented samples, which are generated by applying stochastic augmentations to a given image. We theoretically analyze how our proposed set representation learning can potentially improve the generalization performance at the meta-test. We also empirically validate its effectiveness on various benchmark datasets, showing that Set-SimCLR largely outperforms both UML and instance-level self-supervised learning baselines.

We propose self-supervised set representation learning for unsupervised meta-learning.

We propose Algorithm Distillation (AD), a method for distilling reinforcement learning (RL) algorithms into neural networks by modeling their training histories with a causal sequence model. Algorithm Distillation treats learning to reinforcement learn as an across-episode sequential prediction problem. A dataset of learning histories is generated by a source RL algorithm, and then a causal transformer is trained by autoregressively predicting actions given their preceding learning histories as context. Unlike sequential policy prediction architectures that distill post-learning or expert sequences, AD is able to improve its policy entirely in-context without updating its network parameters. We demonstrate that AD can reinforcement learn in-context in a variety of environments with sparse rewards, combinatorial task structure, and pixel-based observations, and find that AD learns a more data-efficient RL algorithm than the one that generated the source data.

We present Algorithm Distillation, a method that outputs an in-context RL algorithm by treating learning to reinforcement learn as a sequential prediction problem.

The focus on equilibrium solutions in games underemphasizes the importance of understanding their overall structure. A different set of tools is needed for learning and representing the general structure of a game. In this paper we illustrate "Principle Trade-off Analysis" (PTA), a decomposition method that embeds games into a low dimensional feature space and argue that the embeddings are more revealing than previously demonstrated. Here, we develop an analogy to Principal Component Analysis (PCA). PTA represents an arbitrary two-player zero-sum game as the weighted sum of pairs of orthogonal 2D feature planes. We show that each of the feature planes represent unique strategic trade-offs (cyclic modes) and truncation of the sequence provides insightful model reduction. We demonstrate the validity of PTA on a pair of games (Blotto, Pokemon). In Blotto, PTA identifies game symmetries, and specifies strategic trade-offs associated with distinct win conditions. These symmetries reveal limitations of PTA unaddressed in previous work. For Pokemon, PTA recovers clusters that naturally correspond to Pokemon types, correctly identifies the designed tradeoff between those types, and discovers a rock-paper-scissor (RPS) cycle in the Pokemon generation type - all absent any specific information except game outcomes.

A decomposition method that represents a game as a sum of planar embeddings

Topological data analysis (TDA) is a new area of geometric data analysis that focuses on using invariants from algebraic topology to provide multiscale shape descriptors for point clouds. One of the most important shape descriptors is persistent homology, which studies the topological variations as a filtration parameter changes; a typical parameter is the feature scale. For many data sets, it is useful to consider varying multiple filtration parameters at once, for example scale and density. While the theoretical properties of one-parameter persistent homology are well understood, less is known about the multiparameter case. Of particular interest is the problem of representing multiparameter persistent homology by elements of a vector space for integration with traditional machine learning. Existing approaches to this problem either ignore most of the multiparameter information to reduce to the one-parameter case or are heuristic and potentially unstable in the face of noise. In this article, we introduce a general representation framework for multiparameter persistent homology that encompasses previous approaches. We establish theoretical stability guarantees under this framework as well as efficient algorithms for practical computation, making this framework an applicable and versatile tool for TDA practitioners. We validate our stability results and algorithms with numerical experiments that demonstrate statistical convergence, prediction accuracy, and fast running times on several real data sets.

In this article, we provide a general framework for representing multiparameter persistent homology with stability guarantees.

Recent symbolic music generative models have achieved significant improvements in the quality of the generated samples. Nevertheless, it remains hard for users to control the output in such a way that it matches their expectation. To address this limitation, high-level, human-interpretable conditioning is essential. In this work, we release FIGARO, a Transformer-based conditional model trained to generate symbolic music based on a sequence of high-level control codes. To this end, we propose description-to-sequence learning, which consists of automatically extracting fine-grained, human-interpretable features (the description) and training a sequence-to-sequence model to reconstruct the original sequence given only the description as input. FIGARO achieves state-of-the-art performance in multi-track symbolic music generation both in terms of style transfer and sample quality. We show that performance can be further improved by combining human-interpretable with learned features. Our extensive experimental evaluation shows that FIGARO is able to generate samples that closely adhere to the content of the input descriptions, even when they deviate significantly from the training distribution.

We achieve state-of-the-art results in symbolic music style transfer by enabling human-interpretable control over the generation process while improving sample quality at the same time.

We give a complete characterization of the implicit bias of infinitesimal weight decay in the modest setting of univariate one layer ReLU networks. Our main result is a surprisingly simple geometric description of all one layer ReLU networks that exactly fit a dataset $\mathcal D= \set{(x_i,y_i)}$ with the minimum value of the $\ell_2$-norm of the neuron weights. Specifically, we prove that such functions must be either concave or convex between any two consecutive data sites $x_i$ and $x_{i+1}$. Our description implies that interpolating ReLU networks with weak $\ell_2$-regularization achieve the best possible generalization for learning $1d$ Lipschitz functions, up to universal constants.

Minimal \ell_2-norm interpolation by univariate scalar one layer ReLU is completely characterized in terms of the convexity of the learned predictor, giving new sharp generalization bounds on 1d Lipschitz functions.

Gradient regularization (GR) is a method that penalizes the gradient norm of the training loss during training. Although some studies have reported that GR improves generalization performance in deep learning, little attention has been paid to it from the algorithmic perspective, that is, the algorithms of GR that efficiently improve performance. In this study, we first reveal that a specific finite-difference computation, composed of both gradient ascent and descent steps, reduces the computational cost for GR. In addition, this computation empirically achieves better generalization performance. Next, we theoretically analyze a solvable model, a diagonal linear network, and clarify that GR has a desirable implicit bias. Learning with GR chooses better minima in a certain problem, and the finite-difference GR chooses even better ones as the ascent step size becomes larger. Finally, we demonstrate that finite-difference GR is closely related to some other algorithms based on iterative ascent and descent steps for exploring flat minima: sharpness-aware minimization and the flooding method. In particular, we reveal that flooding performs finite-difference GR in an implicit way. Thus, this work broadens our understanding of GR in both practice and theory.

Gradient Regularzation works efficiently by a certain finite-difference computation and has a desirable implicit bias in theory

While the difficulty of reinforcement learning problems is typically related to the complexity of their state spaces, Abstraction proposes that solutions often lie in simpler underlying latent spaces. Prior works have focused on learning either a continuous or dense abstraction, or require a human to provide one. Information-dense representations capture features irrelevant for solving tasks, and continuous spaces can struggle to represent discrete objects. In this work we automatically learn a sparse discrete abstraction of the underlying environment. We do so using a simple end-to-end trainable model based on the successor representation and max-entropy regularization. We describe an algorithm to apply our model, named Discrete State-Action Abstraction (DSAA), which computes an action abstraction in the form of temporally extended actions, i.e., Options, to transition between discrete abstract states. Empirically, we demonstrate the effects of different exploration schemes on our resulting abstraction, and show that it is efficient for solving downstream tasks.

We give a max-entropy regularized model for clustering states based on their successor representation, then train options to navigate between clusters.

No existing spherical convolutional neural network (CNN) framework is both computationally scalable and rotationally equivariant. Continuous approaches capture rotational equivariance but are often prohibitively computationally demanding. Discrete approaches offer more favorable computational performance but at the cost of equivariance. We develop a hybrid discrete-continuous (DISCO) group convolution that is simultaneously equivariant and computationally scalable to high-resolution. While our framework can be applied to any compact group, we specialize to the sphere. Our DISCO spherical convolutions exhibit $\text{SO}(3)$ rotational equivariance, where $\text{SO}(n)$ is the special orthogonal group representing rotations in $n$-dimensions. When restricting rotations of the convolution to the quotient space $\text{SO}(3)/\text{SO}(2)$ for further computational enhancements, we recover a form of asymptotic $\text{SO}(3)$ rotational equivariance. Through a sparse tensor implementation we achieve linear scaling in number of pixels on the sphere for both computational cost and memory usage. For 4k spherical images we realize a saving of $10^9$ in computational cost and $10^4$ in memory usage when compared to the most efficient alternative equivariant spherical convolution. We apply the DISCO spherical CNN framework to a number of benchmark dense-prediction problems on the sphere, such as semantic segmentation and depth estimation, on all of which we achieve the state-of-the-art performance.

A discrete-continuous (DISCO) spherical CNN framework that is simultaneously rotationally equivariant and computationally scalable and achieves state-of-the-art on numerous benchmarks