ICLR 2023 Papers

A number of competing hypotheses have been proposed to explain why small-batch Stochastic Gradient Descent (SGD) leads to improved generalization over the full-batch regime, with recent work crediting the implicit regularization of various quantities throughout training. However, to date, empirical evidence assessing the explanatory power of these hypotheses is lacking. In this paper, we conduct an extensive empirical evaluation, focusing on the ability of various theorized mechanisms to close the small-to-large batch generalization gap. Additionally, we characterize how the quantities that SGD has been claimed to (implicitly) regularize change over the course of training. By using micro-batches, i.e. disjoint smaller subsets of each mini-batch, we empirically show that explicitly penalizing the gradient norm or the Fisher Information Matrix trace, averaged over micro-batches, in the large-batch regime recovers small-batch SGD generalization, whereas Jacobian-based regularizations fail to do so. This generalization performance is shown to often be correlated with how well the regularized model’s gradient norms resemble those of small-batch SGD. We additionally show that this behavior breaks down as the micro-batch size approaches the batch size. Finally, we note that in this line of inquiry, positive experimental findings on CIFAR10 are often reversed on other datasets like CIFAR100, highlighting the need to test hypotheses on a wider collection of datasets.

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.

Recent advancements in explainable machine learning provide effective and faithful solutions for interpreting model behaviors. However, many explanation methods encounter efficiency issues, which largely limit their deployments in practical scenarios. Real-time explainer (RTX) frameworks have thus been proposed to accelerate the model explanation process by learning an one-feed-forward explainer. Existing RTX frameworks typically build the explainer under the supervised learning paradigm, which requires large amounts of explanation labels as the ground truth. Considering that accurate explanation labels are usually hard to obtain, due to constrained computational resources and limited human efforts, effective explainer training is still challenging in practice. In this work, we propose a COntrastive Real-Time eXplanation (CoRTX) framework to learn the explanation-oriented representation and relieve the intensive dependence of explainer training on explanation labels. Specifically, we design a synthetic strategy to select positive and negative instances for explanation representation learning. Theoretical analysis show that our selection strategy can benefit the contrastive learning process on explanation tasks. Experimental results on three real-world datasets further demonstrate the efficiency and efficacy of our proposed CoRTX framework.

Neural Ordinary Differential Equations (NODEs) are a novel neural architecture, built around initial value problems with learned dynamics which are solved during inference. Thought to be inherently more robust against adversarial perturbations, they were recently shown to be vulnerable to strong adversarial attacks, highlighting the need for formal guarantees. However, despite significant progress in robustness verification for standard feed-forward architectures, the verification of high dimensional NODEs remains an open problem. In this work we address this challenge and propose GAINS, an analysis framework for NODEs combining three key ideas: (i) a novel class of ODE solvers, based on variable but discrete time steps, (ii) an efficient graph representation of solver trajectories, and (iii) a novel abstraction algorithm operating on this graph representation. Together, these advances enable the efficient analysis and certified training of high-dimensional NODEs, by reducing the runtime from an intractable $\mathcal{O}(\exp(d)+\exp(T))$ to $\mathcal{O}(d+T^2\log^2T)$ in the dimensionality $d$ and integration time $T$. In an extensive evaluation on computer vision (MNIST and Fashion-MNIST) and time-series forecasting (Physio-Net) problems, we demonstrate the effectiveness of both our certified training and verification methods.

Reparameterization aims to improve the generalization of deep neural networks by transforming a convolution operation into equivalent multi-branched structures during training. However, there exists a gap in understanding how reparameterization may change and benefit learning processes for neural networks. In this paper, we present a novel spatial gradient scaling method to redistribute learning focus among weights in convolutional neural networks. We prove that spatial gradient scaling achieves the same learning dynamics as a branched reparameterization yet without introducing structural changes into the network. We further propose an analytical approach that dynamically learns scalings for each convolutional layer based on the spatial characteristics of its input feature map gauged by mutual information. Experiments on CIFAR-10, CIFAR-100, and ImageNet show that without searching for reparameterized structures, our proposed scaling method outperforms the state-of-the-art reparameterization methods at a lower computational cost.

A standard hardware bottleneck when training deep neural networks is GPU memory. The bulk of memory is occupied by caching intermediate tensors for gradient computation in the backward pass. We propose a novel method to reduce this footprint - Dropping Intermediate Tensors (DropIT). DropIT drops min-k elements of the intermediate tensors and approximates gradients from the sparsified tensors in the backward pass. Theoretically, DropIT reduces noise on estimated gradients and therefore has a higher rate of convergence than vanilla-SGD. Experiments show that we can drop up to 90\% of the intermediate tensor elements in fully-connected and convolutional layers while achieving higher testing accuracy for Visual Transformers and Convolutional Neural Networks on various tasks (e.g., classification, object detection, instance segmentation). Our code and models are available at https://github.com/chenjoya/dropit.

The conventional few-shot classification aims at learning a model on a large labeled base dataset and rapidly adapting to a target dataset that is from the same distribution as the base dataset. However, in practice, the base and the target datasets of few-shot classification are usually from different domains, which is the problem of cross-domain few-shot classification. We tackle this problem by making a small proportion of unlabeled images in the target domain accessible in the training stage. In this setup, even though the base data are sufficient and labeled, the large domain shift still makes transferring the knowledge from the base dataset difficult. We meticulously design a cross-level knowledge distillation method, which can strengthen the ability of the model to extract more discriminative features in the target dataset by guiding the network's shallow layers to learn higher-level information. Furthermore, in order to alleviate the overfitting in the evaluation stage, we propose a feature denoising operation which can reduce the feature redundancy and mitigate overfitting. Our approach can surpass the previous state-of-the-art method, Dynamic-Distillation, by 5.44% on 1-shot and 1.37% on 5-shot classification tasks on average in the BSCD-FSL benchmark. The implementation code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/CLDFD.

Out-of-Distribution (OOD) detection is essential for safety-critical applications of deep neural networks. OOD detection is challenging since DNN models may produce very high logits value even for OOD samples. Hence, it is of great difficulty to discriminate OOD data by directly adopting Softmax on output logits as the confidence score. Differently, we detect the OOD sample with Hopfield energy in a store-then-compare paradigm. In more detail, penultimate layer outputs on the training set are considered as the representations of in-distribution (ID) data. Thus they can be transformed into stored patterns that serve as anchors to measure the discrepancy of unseen data for OOD detection. Starting from the energy function defined in Modern Hopfield Network for the discrepancy score calculation, we derive a simplified version SHE with theoretical analysis. In SHE, we utilize only one stored pattern to present each class, and these patterns can be obtained by simply averaging the penultimate layer outputs of training samples within this class. SHE has the advantages of hyperparameterfree and high computational efficiency. The evaluations of nine widely-used OOD datasets show the promising performance of such a simple yet effective approach and its superiority over State-of-the-Art models. Code is available at https://github.com/zjs975584714/SHE ood detection.

The goal of multi-objective reinforcement learning (MORL) is to learn policies that simultaneously optimize multiple competing objectives. In practice, an agent's preferences over the objectives may not be known apriori, and hence, we require policies that can generalize to arbitrary preferences at test time. In this work, we propose a new data-driven setup for offline MORL, where we wish to learn a preference-agnostic policy agent using only a finite dataset of offline demonstrations of other agents and their preferences. The key contributions of this work are two-fold. First, we introduce D4MORL, (D)atasets for MORL that are specifically designed for offline settings. It contains 1.8 million annotated demonstrations obtained by rolling out reference policies that optimize for randomly sampled preferences on 6 MuJoCo environments with 2-3 objectives each. Second, we propose Pareto-Efficient Decision Agents (PEDA), a family of offline MORL algorithms that builds and extends Decision Transformers via a novel preference-and-return-conditioned policy. Empirically, we show that PEDA closely approximates the behavioral policy on the D4MORL benchmark and provides an excellent approximation of the Pareto-front with appropriate conditioning, as measured by the hypervolume and sparsity metrics.

The approximate nearest neighbor (ANN) search problem is fundamental to efficiently serving many real-world machine learning applications. A number of techniques have been developed for ANN search that are efficient, accurate, and scalable. However, such techniques typically have a number of parameters that affect the speed-recall tradeoff, and exhibit poor performance when such parameters aren't properly set. Tuning these parameters has traditionally been a manual process, demanding in-depth knowledge of the underlying search algorithm. This is becoming an increasingly unrealistic demand as ANN search grows in popularity. To tackle this obstacle to ANN adoption, this work proposes a constrained optimization-based approach to tuning quantization-based ANN algorithms. Our technique takes just a desired search cost or recall as input, and then generates tunings that, empirically, are very close to the speed-recall Pareto frontier and give leading performance on standard benchmarks.

Federated Learning (FL) is a setting for training machine learning models in distributed environments where the clients do not share their raw data but instead send model updates to a server. However, model updates can be subject to attacks and leak private information. Differential Privacy (DP) is a leading mitigation strategy which involves adding noise to clipped model updates, trading off performance for strong theoretical privacy guarantees. Previous work has shown that the threat model of DP is conservative and that the obtained guarantees may be vacuous or may overestimate information leakage in practice. In this paper, we aim to achieve a tighter measurement of the model exposure by considering a realistic threat model. We propose a novel method, CANIFE, that uses canaries - carefully crafted samples by a strong adversary to evaluate the empirical privacy of a training round. We apply this attack to vision models trained on CIFAR-10 and CelebA and to language models trained on Sent140 and Shakespeare. In particular, in realistic FL scenarios, we demonstrate that the empirical per-round epsilon obtained with CANIFE is 4 -- 5$\times$ lower than the theoretical bound.

Statistical inference under market equilibrium effects has attracted increasing attention recently. In this paper we focus on the specific case of linear Fisher markets. They have been widely use in fair resource allocation of food/blood donations and budget management in large-scale Internet ad auctions. In resource allocation, it is crucial to quantify the variability of the resource received by the agents (such as blood banks and food banks) in addition to fairness and efficiency properties of the systems. For ad auction markets, it is important to establish statistical properties of the platform's revenues in addition to their expected values. To this end, we propose a statistical framework based on the concept of infinite-dimensional Fisher markets. In our framework, we observe a market formed by a finite number of items sampled from an underlying distribution (the ``observed market'') and aim to infer several important equilibrium quantities of the underlying long-run market. These equilibrium quantities include individual utilities, social welfare, and pacing multipliers. Through the lens of sample average approximation (SSA), we derive a collection of statistical results and show that the observed market provides useful statistical information of the long-run market. In other words, the equilibrium quantities of the observed market converge to the true ones of the long-run market with strong statistical guarantees. These include consistency, finite sample bounds, asymptotics, and confidence. As an extension, we discuss revenue inference in quasilinear Fisher markets.

Although sparse training has been successfully used in various deep learning tasks to save memory and reduce inference time, the reliability of the produced sparse models remains unexplored. Previous research has shown that deep neural networks tend to be over-confident, and we find that sparse training exacerbates this problem. Therefore, calibrating the sparse models is crucial for reliable prediction and decision making. In this paper, we propose a new sparse training method to produce sparse models with improved confidence calibration. In contrast to previous research that uses only one mask to control the sparse topology, our method utilizes two masks, including a deterministic mask and a random mask. The former efficiently searches and activates important weights by exploiting the magnitude of weights and gradients. While the latter brings better exploration and finds more appropriate weight values by random updates. Theoretically, we prove our method can be viewed as a hierarchical variational approximation of a probabilistic deep Gaussian process. Extensive experiments on multiple datasets, model architectures, and sparsities show that our method can reduce ECE values by up to 47.8\% and simultaneously maintain or even improve accuracy with only a slight increase in computational and storage burden.

We propose Adaptive Deep Kernel Fitting with Implicit Function Theorem (ADKF-IFT), a novel framework for learning deep kernel Gaussian processes (GPs) by interpolating between meta-learning and conventional deep kernel learning. Our approach employs a bilevel optimization objective where we meta-learn generally useful feature representations across tasks, in the sense that task-specific GP models estimated on top of such features achieve the lowest possible predictive loss on average. We solve the resulting nested optimization problem using the implicit function theorem (IFT). We show that our ADKF-IFT framework contains previously proposed Deep Kernel Learning (DKL) and Deep Kernel Transfer (DKT) as special cases. Although ADKF-IFT is a completely general method, we argue that it is especially well-suited for drug discovery problems and demonstrate that it significantly outperforms previous state-of-the-art methods on a variety of real-world few-shot molecular property prediction tasks and out-of-domain molecular property prediction and optimization tasks.

Video scene graph generation (VidSGG) aims to generate a sequence of graph-structure representations for the given video. However, all existing VidSGG methods are fully-supervised, i.e., they need dense and costly manual annotations. In this paper, we propose the first weakly-supervised VidSGG task with only single-frame weak supervision: SF-VidSGG. By ``weakly-supervised", we mean that SF-VidSGG relaxes the training supervision from two different levels: 1) It only provides single-frame annotations instead of all-frame annotations. 2) The single-frame ground-truth annotation is still a weak image SGG annotation, i.e., an unlocalized scene graph. To solve this new task, we also propose a novel Pseudo Label Assignment based method, dubbed as PLA. PLA is a two-stage method, which generates pseudo visual relation annotations for the given video at the first stage, and then trains a fully-supervised VidSGG model with these pseudo labels. Specifically, PLA consists of three modules: an object PLA module, a predicate PLA module, and a future predicate prediction (FPP) module. Firstly, in the object PLA, we localize all objects for every frame. Then, in the predicate PLA, we design two different teachers to assign pseudo predicate labels. Lastly, in the FPP module, we fusion these two predicate pseudo labels by the regularity of relation transition in videos. Extensive ablations and results on the benchmark Action Genome have demonstrated the effectiveness of our PLA.

The idea of embedding optimization problems into deep neural networks as optimization layers to encode constraints and inductive priors has taken hold in recent years. Most existing methods focus on implicitly differentiating Karush–Kuhn–Tucker (KKT) conditions in a way that requires expensive computations on the Jacobian matrix, which can be slow and memory-intensive. In this paper, we developed a new framework, named Alternating Differentiation (Alt-Diff), that differentiates optimization problems (here, specifically in the form of convex optimization problems with polyhedral constraints) in a fast and recursive way. Alt-Diff decouples the differentiation procedure into a primal update and a dual update in an alternating way. Accordingly, Alt-Diff substantially decreases the dimensions of the Jacobian matrix especially for optimization with large-scale constraints and thus increases the computational speed of implicit differentiation. We show that the gradients obtained by Alt-Diff are consistent with those obtained by differentiating KKT conditions. In addition, we propose to truncate Alt-Diff to further accelerate the computational speed. Under some standard assumptions, we show that the truncation error of gradients is upper bounded by the same order of variables' estimation error. Therefore, Alt-Diff can be truncated to further increase computational speed without sacrificing much accuracy. A series of comprehensive experiments validate the superiority of Alt-Diff.

Equipping predicted segmentation with calibrated uncertainty is essential for safety-critical applications. In this work, we focus on capturing the data-inherent uncertainty (aka aleatoric uncertainty) in segmentation, typically when ambiguities exist in input images. Due to the high-dimensional output space and potential multiple modes in segmenting ambiguous images, it remains challenging to predict well-calibrated uncertainty for segmentation. To tackle this problem, we propose a novel mixture of stochastic experts (MoSE) model, where each expert network estimates a distinct mode of the aleatoric uncertainty and a gating network predicts the probabilities of an input image being segmented in those modes. This yields an efficient two-level uncertainty representation. To learn the model, we develop a Wasserstein-like loss that directly minimizes the distribution distance between the MoSE and ground truth annotations. The loss can easily integrate traditional segmentation quality measures and be efficiently optimized via constraint relaxation. We validate our method on the LIDC-IDRI dataset and a modified multimodal Cityscapes dataset. Results demonstrate that our method achieves the state-of-the-art or competitive performance on all metrics.

Reinforcement Learning (RL) algorithms can solve challenging control problems directly from image observations, but they often require millions of environment interactions to do so. Recently, model-based RL algorithms have greatly improved sample-efficiency by concurrently learning an internal model of the world, and supplementing real environment interactions with imagined rollouts for policy improvement. However, learning an effective model of the world from scratch is challenging, and in stark contrast to humans that rely heavily on world understanding and visual cues for learning new skills. In this work, we investigate whether internal models learned by modern model-based RL algorithms can be leveraged to solve new, distinctly different tasks faster. We propose Model-Based Cross-Task Transfer (XTRA), a framework for sample-efficient online RL with scalable pretraining and finetuning of learned world models. By offline multi-task pretraining and online cross-task finetuning, we achieve substantial improvements over a baseline trained from scratch; we improve mean performance of model-based algorithm EfficientZero by 23%, and by as much as 71% in some instances. Project page: https://nicklashansen.github.io/xtra

Image copy detection and retrieval from large databases leverage two components. First, a neural network maps an image to a vector representation, that is relatively robust to various transformations of the image. Second, an efficient but approximate similarity search algorithm trades scalability (size and speed) against quality of the search, thereby introducing a source of error. This paper improves the robustness of image copy detection with active indexing, that optimizes the interplay of these two components. We reduce the quantization loss of a given image representation by making imperceptible changes to the image before its release. The loss is back-propagated through the deep neural network back to the image, under perceptual constraints. These modifications make the image more retrievable. Our experiments show that the retrieval and copy detection of activated images is significantly improved. For instance, activation improves by $+40\%$ the Recall1@1 on various image transformations, and for several popular indexing structures based on product quantization and locality sensitivity hashing.