ICML 2023 论文列表

A directed acyclic graph (DAG) provides valuable prior knowledge that is often discarded in regression tasks in machine learning. We show that the independences arising from the presence of collider structures in DAGs provide meaningful inductive biases, which constrain the regression hypothesis space and improve predictive performance. We introduce collider regression, a framework to incorporate probabilistic causal knowledge from a collider in a regression problem. When the hypothesis space is a reproducing kernel Hilbert space, we prove a strictly positive generalisation benefit under mild assumptions and provide closed-form estimators of the empirical risk minimiser. Experiments on synthetic and climate model data demonstrate performance gains of the proposed methodology.

Test-time adaptation (TTA) adapts a source model to the distribution shift in testing data without using any source data. There have been plenty of algorithms concentrated on covariate shift in the last decade, i.e., $\mathcal{D}_t(X)$, the distribution of the test data is different from the source data. Nonetheless, in real application scenarios, it is necessary to consider the influence of label distribution shift, i.e., both $\mathcal{D}_t(X)$ and $\mathcal{D}_t(Y)$ are shifted, which has not been sufficiently explored yet. To remedy this, we study a new problem setup, namely, TTA with Open-world Data Shift (AODS). The goal of AODS is simultaneously adapting a model to covariate and label distribution shifts in the test phase. In this paper, we first analyze the relationship between classification error and distribution shifts. Motivated by this, we hence propose a new framework, namely ODS, which decouples the mixed distribution shift and then addresses covariate and label distribution shifts accordingly. We conduct experiments on multiple benchmarks with different types of shifts, and the results demonstrate the superior performance of our method against the state of the arts. Moreover, ODS is suitable for many TTA algorithms.

Text-to-image synthesis has recently seen significant progress thanks to large pretrained language models, large-scale training data, and the introduction of scalable model families such as diffusion and autoregressive models. However, the best-performing models require iterative evaluation to generate a single sample. In contrast, generative adversarial networks (GANs) only need a single forward pass. They are thus much faster, but they currently remain far behind the state-of-the-art in large-scale text-to-image synthesis. This paper aims to identify the necessary steps to regain competitiveness. Our proposed model, StyleGAN-T, addresses the specific requirements of large-scale text-to-image synthesis, such as large capacity, stable training on diverse datasets, strong text alignment, and controllable variation vs. text alignment tradeoff. StyleGAN-T significantly improves over previous GANs and outperforms distilled diffusion models - the previous state-of-the-art in fast text-to-image synthesis - in terms of sample quality and speed.

The goal of *data attribution* is to trace model predictions back to training data. Despite a long line of work towards this goal, existing approaches to data attribution tend to force users to choose between computational tractability and efficacy. That is, computationally tractable methods can struggle with accurately attributing model predictions in non-convex settings (e.g., in the context of deep neural networks), while methods that are effective in such regimes require training thousands of models, which makes them impractical for large models or datasets. In this work, we introduce TRAK (Tracing with the Randomly-projected After Kernel), a data attribution method that is both effective *and* computationally tractable for large-scale, differentiable models. In particular, by leveraging only a handful of trained models, TRAK can match the performance of attribution methods that require training thousands of models. We demonstrate the utility of TRAK across various modalities and scales: image classifiers trained on ImageNet, vision-language models (CLIP), and language models (BERT and mT5). We provide code for using TRAK (and reproducing our work) at https://github.com/MadryLab/trak .

Passive observational data, such as human videos, is abundant and rich in information, yet remains largely untapped by current RL methods. Perhaps surprisingly, we show that passive data, despite not having reward or action labels, can still be used to learn features that accelerate downstream RL. Our approach learns from passive data by modeling intentions: measuring how the likelihood of future outcomes change when the agent acts to achieve a particular task. We propose a temporal difference learning objective to learn about intentions, resulting in an algorithm similar to conventional RL, but which learns entirely from passive data. When optimizing this objective, our agent simultaneously learns representations of states, of policies, and of possible outcomes in an environment, all from raw observational data. Both theoretically and empirically, this scheme learns features amenable for value prediction for downstream tasks, and our experiments demonstrate the ability to learn from many forms of passive data, including cross-embodiment video data and YouTube videos.

While efficient distribution learning is no doubt behind the groundbreaking success of diffusion modeling, its theoretical guarantees are quite limited. In this paper, we provide the first rigorous analysis on approximation and generalization abilities of diffusion modeling for well-known function spaces. The highlight of this paper is that when the true density function belongs to the Besov space and the empirical score matching loss is properly minimized, the generated data distribution achieves the nearly minimax optimal estimation rates in the total variation distance and in the Wasserstein distance of order one. Furthermore, we extend our theory to demonstrate how diffusion models adapt to low-dimensional data distributions. We expect these results advance theoretical understandings of diffusion modeling and its ability to generate verisimilar outputs.

We study the efficient estimation of predictive confidence intervals for black-box predictors when the common data exchangeability (e.g., i.i.d.) assumption is violated due to potentially feedback-induced shifts in the input data distribution. That is, we focus on standard and feedback covariate shift (FCS), where the latter allows for feedback dependencies between train and test data that occur in many decision-making scenarios like experimental design. Whereas prior conformal prediction methods for this problem are in general either extremely computationally demanding or make inefficient use of labeled data, we propose a collection of methods based on the jackknife+ that achieve a practical balance of computational and statistical efficiency. Theoretically, our proposed JAW-FCS method extends the rigorous, finite-sample coverage guarantee of the jackknife+ to FCS. We moreover propose two tunable relaxations to JAW-FCS's computation that maintain finite-sample guarantees: one using only $K$ leave-one-out models (JAW-$K$LOO) and a second building on $K$-fold cross validation+ (WCV+). Practically, we demonstrate that JAW-FCS and its computational relaxations outperform state-of-the-art baselines on a variety of real-world datasets under standard and feedback covariate shift, including for biomolecular design and active learning tasks.

Imperfect information games (IIG) are games in which each player only partially observes the current game state. We study how to learn $\epsilon$-optimal strategies in a zero-sum IIG through self-play with trajectory feedback. We give a problem-independent lower bound $\widetilde{\mathcal{O}}(H(A_{\mathcal{X}}+B_{\mathcal{Y}})/\epsilon^2)$ on the required number of realizations to learn these strategies with high probability, where $H$ is the length of the game, $A_{\mathcal{X}}$ and $B_{\mathcal{Y}}$ are the total number of actions for the two players. We also propose two Follow the Regularized leader (FTRL) algorithms for this setting: Balanced FTRL which matches this lower bound, but requires the knowledge of the information set structure beforehand to define the regularization; and Adaptive FTRL which needs $\widetilde{\mathcal{O}}(H^2(A_{\mathcal{X}}+B_{\mathcal{Y}})/\epsilon^2)$ realizations without this requirement by progressively adapting the regularization to the observations.

Although the variational autoencoder (VAE) represents a widely-used deep generative model, the underlying energy function when applied to continuous data remains poorly understood. In fact, most prior theoretical analysis has assumed a simplified affine decoder such that the model collapses to probabilistic PCA, a restricted regime whereby existing classical algorithms can also be trivially applied to guarantee globally optimal solutions. To push our understanding into more complex, practically-relevant settings, this paper instead adopts a deceptively sophisticated single-layer decoder that nonetheless allows the VAE to address the fundamental challenge of learning optimally sparse representations of continuous data originating from popular multiple-response regression models. In doing so, we can then examine VAE properties within the non-trivial context of solving difficult, NP-hard inverse problems. More specifically, we prove rigorous conditions which guarantee that any minimum of the VAE energy (local or global) will produce the optimally sparse latent representation, meaning zero reconstruction error using a minimal number of active latent dimensions. This is ultimately possible because VAE marginalization over the latent posterior selectively smooths away bad local minima as has been conjectured but not actually proven in prior work. We then discuss how equivalent-capacity deterministic autoencoders, even with appropriate sparsity-promoting regularization of the latent space, maintain bad local minima that do not correspond with such parsimonious representations. Overall, these results serve to elucidate key properties of the VAE loss surface relative to finding low-dimensional structure in data.

Deep Neural Networks (DNNs) are one of the most powerful tools for prediction, but many of them implicitly assume that the data are statistically independent. However, in the real world, it is common for large-scale data to be clustered with temporal-spatial correlation structures. Variational approaches and integrated likelihood approaches have been proposed to obtain approximate maximum likelihood estimators (MLEs) for correlated data. However, due to the large size of data, they cannot provide exact MLEs. In this study, we propose a new hierarchical likelihood approach to DNNs with correlated random effects for clustered data. By jointly optimizing the the negative h-likelihood loss, we can provide exact MLEs for both mean and dispersion parameters, as well as the best linear unbiased predictors for the random effects. Moreover, the hierarchical likelihood allows a computable procedure for restricted maximum likelihood estimators of dispersion parameters. The proposed two-step algorithm enables online learning for the neural networks, whereas the integrated likelihood cannot decompose like a widely-used loss function in DNNs. The proposed h-likelihood approach offers several advantages, which we demonstrate through numerical studies and real data analyses.

The surprising ability of Large Language Models (LLMs) to perform well on complex reasoning with only few-shot chain-of-thought prompts is believed to emerge only in very large-scale models. We show that such abilities can, in fact, be distilled down from GPT-3.5 (≥ 175B) to T5 variants (≤ 11B). We propose model specialization, to specialize the model’s ability towards a target task. The hypothesis is that large models (commonly viewed as larger than 100B) have strong modeling power such that they can perform a large spectrum of tasks. Small models (commonly viewed as smaller than 10B) have limited model capacity, but if we specialize their capacity towards a target task, the model can achieve decent performance improvements. We use multi-step math reasoning as our testbed because it is a very typical emergent ability. We show two important aspects of model abilities: (1) balancing language model’s performance on multiple tasks is a delicate matter, as improvements on one task may compromise other tasks; (2) yet by intentionally paying the price of decreased generic ability, we can clearly improve across different model scales smaller than 10B towards a specialized multi-step math reasoning ability. We further give comprehensive discussions about important design choices for better generalization, including the data format mixture and the start model checkpoint. We hope our practice and discoveries can serve as an important attempt towards specialized smaller models in the new research paradigm set by LLMs.

Recurrent Neural Networks (RNNs) offer fast inference on long sequences but are hard to optimize and slow to train. Deep state-space models (SSMs) have recently been shown to perform remarkably well on long sequence modeling tasks, and have the added benefits of fast parallelizable training and RNN-like fast inference. However, while SSMs are superficially similar to RNNs, there are important differences that make it unclear where their performance boost over RNNs comes from. We show that careful design of deep RNNs using standard signal propagation arguments can recover the impressive performance of deep SSMs on long-range reasoning tasks, while matching their training speed. To achieve this, we analyze and ablate a series of changes to standard RNNs including linearizing and diagonalizing the recurrence, using better parameterizations and initializations, and ensuring careful normalization of the forward pass. Our results provide new insights on the origins of the impressive performance of deep SSMs, and introduce an RNN block called the Linear Recurrent Unit (or LRU) that matches both their performance on the Long Range Arena benchmark and their computational efficiency.

The scaling of Transformers has driven breakthrough capabilities for language models. At present, the largest large language models (LLMs) contain upwards of 100B parameters. Vision Transformers (ViT) have introduced the same architecture to image and video modelling, but these have not yet been successfully scaled to nearly the same degree; the largest dense ViT contains 4B parameters (Chen et al., 2022). We present a recipe for highly efficient and stable training of a 22B-parameter ViT (ViT-22B) and perform a wide variety of experiments on the resulting model. When evaluated on downstream tasks (often with a lightweight linear model on frozen features), ViT-22B demonstrates increasing performance with scale. We further observe other interesting benefits of scale, including an improved tradeoff between fairness and performance, state-of-the-art alignment to human visual perception in terms of shape/texture bias, and improved robustness. ViT-22B demonstrates the potential for "LLM-like" scaling in vision, and provides key steps towards getting there.

Although Nesterov's accelerated gradient method (AGM) has been studied from various perspectives, it remains unclear why the most popular forms of AGMs must handle convex and strongly convex objective functions separately. To address this inconsistency, we propose a novel unified framework for Lagrangians, ordinary differential equation (ODE) models, and algorithms. As a special case, our new simple momentum algorithm, which we call the unified AGM, seamlessly bridges the gap between the two most popular forms of Nesterov's AGM and has a superior convergence guarantee compared to existing algorithms for non-strongly convex objective functions. This property is beneficial in practice when considering ill-conditioned $\mu$-strongly convex objective functions (with small $\mu$). Furthermore, we generalize this algorithm and the corresponding ODE model to the higher-order non-Euclidean setting. Last but not least, our unified framework is used to construct the unified AGM-G ODE, a novel ODE model for minimizing the gradient norm of strongly convex functions.

We attack the state-of-the-art Go-playing AI system KataGo by training adversarial policies against it, achieving a >97% win rate against KataGo running at superhuman settings. Our adversaries do not win by playing Go well. Instead, they trick KataGo into making serious blunders. Our attack transfers zero-shot to other superhuman Go-playing AIs, and is comprehensible to the extent that human experts can implement it without algorithmic assistance to consistently beat superhuman AIs. The core vulnerability uncovered by our attack persists even in KataGo agents adversarially trained to defend against our attack. Our results demonstrate that even superhuman AI systems may harbor surprising failure modes. Example games are available https://goattack.far.ai/.

Off-policy learning (OPL) aims at finding improved policies from logged bandit data, often by minimizing the inverse propensity scoring (IPS) estimator of the risk. In this work, we investigate a smooth regularization for IPS, for which we derive a two-sided PAC-Bayes generalization bound. The bound is tractable, scalable, interpretable and provides learning certificates. In particular, it is also valid for standard IPS without making the assumption that the importance weights are bounded. We demonstrate the relevance of our approach and its favorable performance through a set of learning tasks. Since our bound holds for standard IPS, we are able to provide insight into when regularizing IPS is useful. Namely, we identify cases where regularization might not be needed. This goes against the belief that, in practice, clipped IPS often enjoys favorable performance than standard IPS in OPL.

The proposed method, **Discriminator Guidance**, aims to improve sample generation of pre-trained diffusion models. The approach introduces a discriminator that gives explicit supervision to a denoising sample path whether it is realistic or not. Unlike GANs, our approach does not require joint training of score and discriminator networks. Instead, we train the discriminator after score training, making discriminator training stable and fast to converge. In sample generation, we add an auxiliary term to the pre-trained score to deceive the discriminator. This term corrects the model score to the data score at the optimal discriminator, which implies that the discriminator helps better score estimation in a complementary way. Using our algorithm, we achive state-of-the-art results on ImageNet 256x256 with FID 1.83 and recall 0.64, similar to the validation data's FID (1.68) and recall (0.66). We release the code at https://github.com/alsdudrla10/DG.

Current self-supervised learning algorithms are often modality-specific and require large amounts of computational resources. To address these issues, we increase the training efficiency of data2vec, a learning objective that generalizes across several modalities. We do not encode masked tokens, use a fast convolutional decoder and amortize the effort to build teacher representations. data2vec 2.0 benefits from the rich contextualized target representations introduced in data2vec which enable a fast self-supervised learner. Experiments on ImageNet-1K image classification show that data2vec 2.0 matches the accuracy of Masked Autoencoders in 16.4x lower pre-training time, on Librispeech speech recognition it performs as well as wav2vec 2.0 in 10.6x less time, and on GLUE natural language understanding it matches a retrained RoBERTa model in half the time. Trading some speed for accuracy results in ImageNet-1K top-1 accuracy of 86.8% with a ViT-L model trained for 150 epochs.

We provide an efficient implementation of the backpropagation algorithm, specialized to the case where the weights of the neural network being trained are _sparse_. Our algorithm is general, as it applies to arbitrary (unstructured) sparsity and common layer types (e.g., convolutional or linear). We provide a fast vectorized implementation on commodity CPUs, and show that it can yield speedups in end-to-end runtime experiments, both in transfer learning using already-sparsified networks, and in training sparse networks from scratch. Thus, our results provide the first support for sparse training on commodity hardware.