NeurIPS 2022 Papers

Constrained Markov game is a fundamental problem that covers many applications, where multiple players compete with each other under behavioral constraints. The existing literature has proved the existence of Nash equilibrium for constrained Markov games, which turns out to be PPAD-complete and cannot be computed in polynomial time. In this work, we propose a surrogate notion of correlated equilibrium (CE) for constrained Markov games that can be computed in polynomial time, and study its fundamental properties. We show that the modification structure of CE of constrained Markov games is fundamentally different from that of unconstrained Markov games. Moreover, we prove that the corresponding Lagrangian function has zero duality gap. Based on this result, we develop the first primal-dual algorithm that provably converges to CE of constrained Markov games. In particular, we prove that both the duality gap and the constraint violation of the output policy converge at the rate $\mathcal{O}(\frac{1}{\sqrt{T}})$. Moreover, when adopting the V-learning algorithm as the subroutine in the primal update, our algorithm achieves an approximate CE with $\epsilon$ duality gap with the sample complexity $\mathcal{O}(H^9|\mathcal{S}||\mathcal{A}|^{2} \epsilon^{-4})$.

Recent self-supervised advances in medical computer vision exploit the global and local anatomical self-similarity for pretraining prior to downstream tasks such as segmentation. However, current methods assume i.i.d. image acquisition, which is invalid in clinical study designs where follow-up longitudinal scans track subject-specific temporal changes. Further, existing self-supervised methods for medically-relevant image-to-image architectures exploit only spatial or temporal self-similarity and do so via a loss applied only at a single image-scale, with naive multi-scale spatiotemporal extensions collapsing to degenerate solutions. To these ends, this paper makes two contributions: (1) It presents a local and multi-scale spatiotemporal representation learning method for image-to-image architectures trained on longitudinal images. It exploits the spatiotemporal self-similarity of learned multi-scale intra-subject image features for pretraining and develops several feature-wise regularizations that avoid degenerate representations; (2) During finetuning, it proposes a surprisingly simple self-supervised segmentation consistency regularization to exploit intra-subject correlation. Benchmarked across various segmentation tasks, the proposed framework outperforms both well-tuned randomly-initialized baselines and current self-supervised techniques designed for both i.i.d. and longitudinal datasets. These improvements are demonstrated across both longitudinal neurodegenerative adult MRI and developing infant brain MRI and yield both higher performance and longitudinal consistency.

We consider the problem of minimizing the number of matrix-vector queries needed for accurate trace estimation in the dynamic setting where our underlying matrix is changing slowly, such as during an optimization process. Specifically, for any $m$ matrices $\mathbf{A}_1,...,\mathbf{A}_m$ with consecutive differences bounded in Schatten-$1$ norm by $\alpha$, we provide a novel binary tree summation procedure that simultaneously estimates all $m$ traces up to $\epsilon$ error with $\delta$ failure probability with an optimal query complexity of $\widetilde{O}(m \alpha\sqrt{\log(1/\delta)}/\epsilon + m\log(1/\delta))$, improving the dependence on both $\alpha$ and $\delta$ from Dharangutte and Musco (NeurIPS, 2021). Our procedure works without additional norm bounds on $\mathbf{A}_i$ and can be generalized to a bound for the $p$-th Schatten norm for $p \in [1,2]$, giving a complexity of $\widetilde{O}(m \alpha(\sqrt{\log(1/\delta)}/\epsilon)^p +m \log(1/\delta))$. By using novel reductions to communication complexity and information-theoretic analyses of Gaussian matrices, we provide matching lower bounds for static and dynamic trace estimation in all relevant parameters, including the failure probability. Our lower bounds (1) give the first tight bounds for Hutchinson's estimator in the matrix-vector product model with Frobenius norm error {\it even in the static setting}, and (2) are the first unconditional lower bounds for dynamic trace estimation, resolving open questions of prior work.

Long-term fairness is an important factor of consideration in designing and deploying learning-based decision systems in high-stake decision-making contexts. Recent work has proposed the use of Markov Decision Processes (MDPs) to formulate decision-making with long-term fairness requirements in dynamically changing environments, and demonstrated major challenges in directly deploying heuristic and rule-based policies that worked well in static environments. We show that policy optimization methods from deep reinforcement learning can be used to find strictly better decision policies that can often achieve both higher overall utility and less violation of the fairness requirements, compared to previously-known strategies. In particular, we propose new methods for imposing fairness requirements in policy optimization by regularizing the advantage evaluation of different actions. Our proposed methods make it easy to impose fairness constraints without reward engineering or sacrificing training efficiency. We perform detailed analyses in three established case studies, including attention allocation in incident monitoring, bank loan approval, and vaccine distribution in population networks.

The prevalence of data scraping from social media as a means to obtain datasets has led to growing concerns regarding unauthorized use of data. Data poisoning attacks have been proposed as a bulwark against scraping, as they make data ``unlearnable'' by adding small, imperceptible perturbations. Unfortunately, existing methods require knowledge of both the target architecture and the complete dataset so that a surrogate network can be trained, the parameters of which are used to generate the attack. In this work, we introduce autoregressive (AR) poisoning, a method that can generate poisoned data without access to the broader dataset. The proposed AR perturbations are generic, can be applied across different datasets, and can poison different architectures. Compared to existing unlearnable methods, our AR poisons are more resistant against common defenses such as adversarial training and strong data augmentations. Our analysis further provides insight into what makes an effective data poison.

Numerical validation is at the core of machine learning research as it allows us to assess the actual impact of new methods, and to confirm the agreement between theory and practice. Yet, the rapid development of the field poses several challenges: researchers are confronted with a profusion of methods to compare, limited transparency and consensus on best practices, as well as tedious re-implementation work. As a result, validation is often very partial, which can lead to wrong conclusions that slow down the progress of research. We propose Benchopt, a collaborative framework to automatize, publish and reproduce optimization benchmarks in machine learning across programming languages and hardware architectures. Benchopt simplifies benchmarking for the community by providing an off-the-shelf tool for running, sharing and extending experiments. To demonstrate its broad usability, we showcase benchmarks on three standard ML tasks: $\ell_2$-regularized logistic regression, Lasso and ResNet18 training for image classification. These benchmarks highlight key practical findings that give a more nuanced view of state-of-the-art for these problems, showing that for practical evaluation, the devil is in the details.

Despite substantial progress in 3D object detection, advanced 3D detectors often suffer from heavy computation overheads. To this end, we explore the potential of knowledge distillation (KD) for developing efficient 3D object detectors, focusing on popular pillar- and voxel-based detectors. In the absence of well-developed teacher-student pairs, we first study how to obtain student models with good trade offs between accuracy and efficiency from the perspectives of model compression and input resolution reduction. Then, we build a benchmark to assess existing KD methods developed in the 2D domain for 3D object detection upon six well-constructed teacher-student pairs. Further, we propose an improved KD pipeline incorporating an enhanced logit KD method that performs KD on only a few pivotal positions determined by teacher classification response and a teacher-guided student model initialization to facilitate transferring teacher model's feature extraction ability to students through weight inheritance. Finally, we conduct extensive experiments on the Waymo dataset. Our best performing model achieves $65.75\%$ LEVEL 2 mAPH surpassing its teacher model and requiring only $44\%$ of teacher flops. Our most efficient model runs 51 FPS on an NVIDIA A100, which is $2.2\times$ faster than PointPillar with even higher accuracy. Code will be available.

Dynamic graph neural networks (DyGNNs) have demonstrated powerful predictive abilities by exploiting graph structural and temporal dynamics. However, the existing DyGNNs fail to handle distribution shifts, which naturally exist in dynamic graphs, mainly because the patterns exploited by DyGNNs may be variant with respect to labels under distribution shifts. In this paper, we propose to handle spatio-temporal distribution shifts in dynamic graphs by discovering and utilizing {\it invariant patterns}, i.e., structures and features whose predictive abilities are stable across distribution shifts, which faces two key challenges: 1) How to discover the complex variant and invariant spatio-temporal patterns in dynamic graphs, which involve both time-varying graph structures and node features. 2) How to handle spatio-temporal distribution shifts with the discovered variant and invariant patterns. To tackle these challenges, we propose the Disentangled Intervention-based Dynamic graph Attention networks (DIDA). Our proposed method can effectively handle spatio-temporal distribution shifts in dynamic graphs by discovering and fully utilizing invariant spatio-temporal patterns. Specifically, we first propose a disentangled spatio-temporal attention network to capture the variant and invariant patterns. Then, we design a spatio-temporal intervention mechanism to create multiple interventional distributions by sampling and reassembling variant patterns across neighborhoods and time stamps to eliminate the spurious impacts of variant patterns. Lastly, we propose an invariance regularization term to minimize the variance of predictions in intervened distributions so that our model can make predictions based on invariant patterns with stable predictive abilities and therefore handle distribution shifts. Experiments on three real-world datasets and one synthetic dataset demonstrate the superiority of our method over state-of-the-art baselines under distribution shifts. Our work is the first study of spatio-temporal distribution shifts in dynamic graphs, to the best of our knowledge.

Neural Ordinary Differential Equations (Neural ODEs) are the continuous analog of Residual Neural Networks (ResNets). We investigate whether the discrete dynamics defined by a ResNet are close to the continuous one of a Neural ODE. We first quantify the distance between the ResNet's hidden state trajectory and the solution of its corresponding Neural ODE. Our bound is tight and, on the negative side, does not go to $0$ with depth $N$ if the residual functions are not smooth with depth. On the positive side, we show that this smoothness is preserved by gradient descent for a ResNet with linear residual functions and small enough initial loss. It ensures an implicit regularization towards a limit Neural ODE at rate $\frac1N$, uniformly with depth and optimization time. As a byproduct of our analysis, we consider the use of a memory-free discrete adjoint method to train a ResNet by recovering the activations on the fly through a backward pass of the network, and show that this method theoretically succeeds at large depth if the residual functions are Lipschitz with the input. We then show that Heun's method, a second order ODE integration scheme, allows for better gradient estimation with the adjoint method when the residual functions are smooth with depth. We experimentally validate that our adjoint method succeeds at large depth, and that Heun’s method needs fewer layers to succeed. We finally use the adjoint method successfully for fine-tuning very deep ResNets without memory consumption in the residual layers.

In recent years, the neural stochastic differential equation (NSDE) has gained attention for modeling stochastic representations with great success in various types of applications. However, it typically loses expressivity when the data representation is manifold-valued. To address this issue, we suggest a principled method for expressing the stochastic representation with the Riemannian neural SDE (RNSDE), which extends the conventional Euclidean NSDE. Empirical results for various tasks demonstrate that the proposed method significantly outperforms baseline methods.

Although recent point cloud analysis achieves impressive progress, the paradigm of representation learning from single modality gradually meets its bottleneck. In this work, we take a step towards more discriminative 3D point cloud representation using 2D images, which inherently contain richer appearance information, e.g., texture, color, and shade. Specifically, this paper introduces a simple but effective point cloud cross-modality training (PointCMT) strategy, which utilizes view-images, i.e., rendered or projected 2D images of the 3D object, to boost point cloud classification. In practice, to effectively acquire auxiliary knowledge from view-images, we develop a teacher-student framework and formulate the cross-modal learning as a knowledge distillation problem. Through novel feature and classifier enhancement criteria, PointCMT eliminates the distribution discrepancy between different modalities and avoid potential negative transfer effectively. Note that PointCMT efficiently improves the point-only representation without any architecture modification. Sufficient experiments verify significant gains on various datasets based on several backbones, i.e., equipped with PointCMT, PointNet++ and PointMLP achieve state-of-the-art performance on two benchmarks, i.e., 94.4% and 86.7% accuracy on ModelNet40 and ScanObjectNN, respectively.

Information-directed sampling (IDS) has revealed its potential as a data-efficient algorithm for reinforcement learning (RL). However, theoretical understanding of IDS for Markov Decision Processes (MDPs) is still limited. We develop novel information-theoretic tools to bound the information ratio and cumulative information gain about the learning target. Our theoretical results shed light on the importance of choosing the learning target such that the practitioners can balance the computation and regret bounds. As a consequence, we derive prior-free Bayesian regret bounds for vanilla-IDS which learns the whole environment under tabular finite-horizon MDPs. In addition, we propose a computationally-efficient regularized-IDS that maximizes an additive form rather than the ratio form and show that it enjoys the same regret bound as vanilla-IDS. With the aid of rate-distortion theory, we improve the regret bound by learning a surrogate, less informative environment. Furthermore, we extend our analysis to linear MDPs and prove similar regret bounds for Thompson sampling as a by-product.

Online imitation learning is the problem of how best to mimic expert demonstrations, given access to the environment or an accurate simulator. Prior work has shown that in the \textit{infinite} sample regime, exact moment matching achieves value equivalence to the expert policy. However, in the \textit{finite} sample regime, even if one has no optimization error, empirical variance can lead to a performance gap that scales with $H^2 / N_{\text{exp}}$ for behavioral cloning and $H / N_{\text{exp}}$ for online moment matching, where $H$ is the horizon and $N_{\text{exp}}$ is the size of the expert dataset. We introduce the technique of ``replay estimation'' to reduce this empirical variance: by repeatedly executing cached expert actions in a stochastic simulator, we compute a smoother expert visitation distribution estimate to match. In the presence of general function approximation, we prove a meta theorem reducing the performance gap of our approach to the \textit{parameter estimation error} for offline classification (i.e. learning the expert policy). In the tabular setting or with linear function approximation, our meta theorem shows that the performance gap incurred by our approach achieves the optimal $\widetilde{O} \left( \min( H^{3/2} / N_{\text{exp}}, H / \sqrt{N_{\text{exp}}} \right)$ dependency, under significantly weaker assumptions compared to prior work. We implement multiple instantiations of our approach on several continuous control tasks and find that we are able to significantly improve policy performance across a variety of dataset sizes.

We consider 1-dimensional location estimation, where we estimate a parameter $\lambda$ from $n$ samples $\lambda + \eta_i$, with each $\eta_i$ drawn i.i.d. from a known distribution $f$. For fixed $f$ the maximum-likelihood estimate (MLE) is well-known to be optimal in the limit as $n \to \infty$: it is asymptotically normal with variance matching the Cramer-Rao lower bound of $\frac{1}{n\mathcal{I}}$, where $\mathcal{I}$ is the Fisher information of $f$. However, this bound does not hold for finite $n$, or when $f$ varies with $n$. We show for arbitrary $f$ and $n$ that one can recover a similar theory based on the Fisher information of a smoothed version of $f$, where the smoothing radius decays with $n$.

Cross-device federated learning is an emerging machine learning (ML) paradigm where a large population of devices collectively train an ML model while the data remains on the devices. This research field has a unique set of practical challenges, and to systematically make advances, new datasets curated to be compatible with this paradigm are needed. Existing federated learning benchmarks in the image domain do not accurately capture the scale and heterogeneity of many real-world use cases. We introduce FLAIR, a challenging large-scale annotated image dataset for multi-label classification suitable for federated learning. FLAIR has 429,078 images from 51,414 Flickr users and captures many of the intricacies typically encountered in federated learning, such as heterogeneous user data and a long-tailed label distribution. We implement multiple baselines in different learning setups for different tasks on this dataset. We believe FLAIR can serve as a challenging benchmark for advancing the state-of-the art in federated learning. Dataset access and the code for the benchmark are available at https://github.com/apple/ml-flair.

Weakly-supervised whole-slide image (WSI) classification (WSWC) is a challenging task where a large number of unlabeled patches (instances) exist within each WSI (bag) while only a slide label is given. Despite recent progress for the multiple instance learning (MIL)-based WSI analysis, the major limitation is that it usually focuses on the easy-to-distinguish diagnosis-positive regions while ignoring positives that occupy a small ratio in the entire WSI. To obtain more discriminative features, we propose a novel weakly-supervised classification method based on cross-slide contrastive learning (called SCL-WC), which depends on task-agnostic self-supervised feature pre-extraction and task-specific weakly-supervised feature refinement and aggregation for WSI-level prediction. To enable both intra-WSI and inter-WSI information interaction, we propose a positive-negative-aware module (PNM) and a weakly-supervised cross-slide contrastive learning (WSCL) module, respectively. The WSCL aims to pull WSIs with the same disease types closer and push different WSIs away. The PNM aims to facilitate the separation of tumor-like patches and normal ones within each WSI. Extensive experiments demonstrate state-of-the-art performance of our method in three different classification tasks (e.g., over 2% of AUC in Camelyon16, 5% of F1 score in BRACS, and 3% of AUC in DiagSet). Our method also shows superior flexibility and scalability in weakly-supervised localization and semi-supervised classification experiments (e.g., first place in the BRIGHT challenge). Our code will be available at https://github.com/Xiyue-Wang/SCL-WC.

Human brains are commonly modeled as networks of Regions of Interest (ROIs) and their connections for the understanding of brain functions and mental disorders. Recently, Transformer-based models have been studied over different types of data, including graphs, shown to bring performance gains widely. In this work, we study Transformer-based models for brain network analysis. Driven by the unique properties of data, we model brain networks as graphs with nodes of fixed size and order, which allows us to (1) use connection profiles as node features to provide natural and low-cost positional information and (2) learn pair-wise connection strengths among ROIs with efficient attention weights across individuals that are predictive towards downstream analysis tasks. Moreover, we propose an Orthonormal Clustering Readout operation based on self-supervised soft clustering and orthonormal projection. This design accounts for the underlying functional modules that determine similar behaviors among groups of ROIs, leading to distinguishable cluster-aware node embeddings and informative graph embeddings. Finally, we re-standardize the evaluation pipeline on the only one publicly available large-scale brain network dataset of ABIDE, to enable meaningful comparison of different models. Experiment results show clear improvements of our proposed Brain Network Transformer on both the public ABIDE and our restricted ABCD datasets. The implementation is available at https://github.com/Wayfear/BrainNetworkTransformer.

The recent success of Transformer has benefited many real-world applications, with its capability of building long dependency through pairwise dot-products. However, the strong assumption that elements are directly attentive to each other limits the performance of tasks with high-order dependencies such as natural language understanding and Image captioning. To solve such problems, we are the first to define the Jump Self-attention (JAT) to build Transformers. Inspired by the pieces moving of English Draughts, we introduce the spectral convolutional technique to calculate JAT on the dot-product feature map. This technique allows JAT's propagation in each self-attention head and is interchangeable with the canonical self-attention. We further develop the higher-order variants under the multi-hop assumption to increase the generality. Moreover, the proposed architecture is compatible with the pre-trained models. With extensive experiments, we empirically show that our methods significantly increase the performance on ten different tasks.

The dynamic membrane potential threshold, as one of the essential properties of a biological neuron, is a spontaneous regulation mechanism that maintains neuronal homeostasis, i.e., the constant overall spiking firing rate of a neuron. As such, the neuron firing rate is regulated by a dynamic spiking threshold, which has been extensively studied in biology. Existing work in the machine learning community does not employ bioinspired spiking threshold schemes. This work aims at bridging this gap by introducing a novel bioinspired dynamic energy-temporal threshold (BDETT) scheme for spiking neural networks (SNNs). The proposed BDETT scheme mirrors two bioplausible observations: a dynamic threshold has 1) a positive correlation with the average membrane potential and 2) a negative correlation with the preceding rate of depolarization. We validate the effectiveness of the proposed BDETT on robot obstacle avoidance and continuous control tasks under both normal conditions and various degraded conditions, including noisy observations, weights, and dynamic environments. We find that the BDETT outperforms existing static and heuristic threshold approaches by significant margins in all tested conditions, and we confirm that the proposed bioinspired dynamic threshold scheme offers homeostasis to SNNs in complex real-world tasks.