ICML 2024 Papers

Opponent exploitation is an important task for players to exploit the weaknesses of others in games. Existing approaches mainly focus on balancing between exploitation and exploitability but are often vulnerable to modeling errors and deceptive adversaries. To address this problem, our paper offers a novel perspective on the safety of opponent exploitation, named Adaptation Safety. This concept leverages the insight that strategies, even those not explicitly aimed at opponent exploitation, may inherently be exploitable due to computational complexities, rendering traditional safety overly rigorous. In contrast, adaptation safety requires that the strategy should not be more exploitable than it would be in scenarios where opponent exploitation is not considered. Building on such adaptation safety, we further propose an Opponent eXploitation Search (OX-Search) framework by incorporating real-time search techniques for efficient online opponent exploitation. Moreover, we provide theoretical analyses to show the adaptation safety and robust exploitation of OX-Search, even with inaccurate opponent models. Empirical evaluations in popular poker games demonstrate OX-Search's superiority in both exploitability and exploitation compared to previous methods.

Despite their stellar performance on a wide range of tasks, including in-context tasks only revealed during inference, vanilla transformers and variants trained for next-token predictions (a) do not learn an explicit world model of their environment which can be flexibly queried and (b) cannot be used for planning or navigation. In this paper, we consider partially observed environments (POEs), where an agent receives perceptually aliased observations as it navigates, which makes path planning hard. We introduce a transformer with (multiple) discrete bottleneck(s), TDB, whose latent codes learn a compressed representation of the history of observations and actions. After training a TDB to predict the future observation(s) given the history, we extract interpretable cognitive maps of the environment from its active bottleneck(s) indices. These maps are then paired with an external solver to solve (constrained) path planning problems. First, we show that a TDB trained on POEs (a) retains the near-perfect predictive performance of a vanilla transformer or an LSTM while (b) solving shortest path problems exponentially faster. Second, a TDB extracts interpretable representations from text datasets, while reaching higher in-context accuracy than vanilla sequence models. Finally, in new POEs, a TDB (a) reaches near-perfect in-context accuracy, (b) learns accurate in-context cognitive maps (c) solves in-context path planning problems.

Unbiased Learning to Rank (ULTR) aims to train unbiased ranking models from biased click logs, by explicitly modeling a generation process for user behavior and fitting click data based on examination hypothesis. Previous research found empirically that the true latent relevance is mostly recoverable through click fitting. However, we demonstrate that this is not always achievable, resulting in a significant reduction in ranking performance. This research investigates the conditions under which relevance can be recovered from click data in the first principle. We initially characterize a ranking model as identifiable if it can recover the true relevance up to a scaling transformation, a criterion sufficient for the pairwise ranking objective. Subsequently, we investigate an equivalent condition for identifiability, articulated as a graph connectivity test problem: the recovery of relevance is feasible if and only if the identifiability graph (IG), derived from the underlying structure of the dataset, is connected. The presence of a disconnected IG may lead to degenerate cases and suboptimal ranking performance. To tackle this challenge, we introduce two methods, namely node intervention and node merging, designed to modify the dataset and restore the connectivity of the IG. Empirical results derived from a simulated dataset and two real-world LTR benchmark datasets not only validate our proposed theory, but also demonstrate the effectiveness of our methods in alleviating data bias when the relevance model is unidentifiable.

The core of clustering lies in incorporating prior knowledge to construct supervision signals. From classic k-means based on data compactness to recent contrastive clustering guided by self-supervision, the evolution of clustering methods intrinsically corresponds to the progression of supervision signals. At present, substantial efforts have been devoted to mining internal supervision signals from data. Nevertheless, the abundant external knowledge such as semantic descriptions, which naturally conduces to clustering, is regrettably overlooked. In this work, we propose leveraging external knowledge as a new supervision signal to guide clustering. To implement and validate our idea, we design an externally guided clustering method (Text-Aided Clustering, TAC), which leverages the textual semantics of WordNet to facilitate image clustering. Specifically, TAC first selects and retrieves WordNet nouns that best distinguish images to enhance the feature discriminability. Then, TAC collaborates text and image modalities by mutually distilling cross-modal neighborhood information. Experiments demonstrate that TAC achieves state-of-the-art performance on five widely used and three more challenging image clustering benchmarks, including the full ImageNet-1K dataset. The code can be accessed at https://github.com/XLearning-SCU/2024-ICML-TAC.

In this paper, we take a step towards a deeper understanding of learning from human preferences by systematically comparing the paradigm of reinforcement learning from human feedback (RLHF) with the recently proposed paradigm of direct preference optimization (DPO). We focus our attention on the class of loglinear policy parametrization and linear reward functions. In order to compare the two paradigms, we first derive minimax statistical bounds on the suboptimality gap induced by both RLHF and DPO, assuming access to an oracle that exactly solves the optimization problems. We provide a detailed discussion on the relative comparison between the two paradigms, simultaneously taking into account the sample size, policy and reward class dimensions, and the regularization temperature. Moreover, we extend our analysis to the approximate optimization setting and derive exponentially decaying convergence rates for both RLHF and DPO. Next, we analyze the setting where the ground-truth reward is not realizable and find that, while RLHF incurs a constant additional error, DPO retains its asymptotically decaying gap by just tuning the temperature accordingly. Finally, we extend our comparison to the Markov decision process setting, where we generalize our results with exact optimization. To the best of our knowledge, we are the first to provide such a comparative analysis for RLHF and DPO.

In our study, we present bifurcated attention, a method developed for language model inference in single-context batch sampling contexts. This approach aims to reduce redundant memory IO costs, a significant factor in latency for high batch sizes and long context lengths. Bifurcated attention achieves this by dividing the attention mechanism during incremental decoding into two distinct GEMM operations, focusing on the KV cache from prefill and the decoding process. This method ensures precise computation and maintains the usual computational load (FLOPs) of standard attention mechanisms, but with reduced memory IO. Bifurcated attention is also compatible with multi-query attention mechanism known for reduced memory IO for KV cache, further enabling higher batch size and context length. The resulting efficiency leads to lower latency, improving suitability for real-time applications, e.g., enabling massively-parallel answer generation without substantially increasing latency, enhancing performance when integrated with post-processing techniques such as reranking.

In this work, we introduce the Prototypical Transformer (ProtoFormer), a general and unified framework that approaches various motion tasks from a prototype perspective. ProtoFormer seamlessly integrates prototype learning with Transformer by thoughtfully considering motion dynamics, introducing two innovative designs. First, Cross-Attention Prototyping discovers prototypes based on signature motion patterns, providing transparency in understanding motion scenes. Second, Latent Synchronization guides feature representation learning via prototypes, effectively mitigating the problem of motion uncertainty. Empirical results demonstrate that our approach achieves competitive performance on popular motion tasks such as optical flow and scene depth. Furthermore, it exhibits generality across various downstream tasks, including object tracking and video stabilization.

Reasoning is a fundamental ability towards artificial general intelligence (AGI). Fueled by the success of deep learning, the neural logic machines models (NLMs) have introduced novel neural-symbolic structures and demonstrate great performance and generalization on reasoning and decision-making tasks. However, the original training approaches of the NLMs are still far from perfect, the models would repeat similar mistakes during the training process which leads to sub-optimal performance. To mitigate this issue, we present a novel framework named Failure Reflection Guided Regularizer (FRGR). FRGR first dynamically identifies and summarizes the root cause if the model repeats similar mistakes during training. Then it penalizes the model if it makes similar mistakes in future training iterations. In this way, the model is expected to avoid repeating errors of similar root causes and converge faster to a better-performed optimum. Experimental results on multiple relational reasoning and decision-making tasks demonstrate the effectiveness of FRGR in improving performance, generalization, training efficiency, and data efficiency.

Policy optimization from batch data, i.e., offline reinforcement learning (RL) is important when collecting data from a current policy is not possible. This setting incurs distribution mismatch between batch training data and trajectories from the current policy. Pessimistic offsets estimate mismatch using concentration bounds, which possess strong theoretical guarantees and simplicity of implementation. Mismatch may be conservative in sparse data regions and less so otherwise, which can result in under-performing their no-penalty variants in practice. We derive a new pessimistic penalty as the distance between the data and the true distribution using an evaluable one-sample test known as Stein Discrepancy that requires minimal smoothness conditions, and noticeably, allows a mixture family representation of distribution over next states. This entity forms a quantifier of information in offline data, which justifies calling this approach *information-directed pessimism* (IDP) for offline RL. We further establish that this new penalty based on discrete Stein discrepancy yields practical gains in performance while generalizing the regret of prior art to multimodal distributions.

This paper describes a differentially private post-processing algorithm for learning fair regressors satisfying statistical parity, addressing privacy concerns of machine learning models trained on sensitive data, as well as fairness concerns of their potential to propagate historical biases. Our algorithm can be applied to post-process any given regressor to improve fairness by remapping its outputs. It consists of three steps: first, the output distributions are estimated privately via histogram density estimation and the Laplace mechanism, then their Wasserstein barycenter is computed, and the optimal transports to the barycenter are used for post-processing to satisfy fairness. We analyze the sample complexity of our algorithm and provide fairness guarantee, revealing a trade-off between the statistical bias and variance induced from the choice of the number of bins in the histogram, in which using less bins always favors fairness at the expense of error.

Estimating the uncertainty of a model’s prediction on a test point is a crucial part of ensuring reliability and calibration under distribution shifts.A minimum description length approach to this problem uses the predictive normalized maximum likelihood (pNML) distribution, which considers every possible label for a data point, and decreases confidence in a prediction if other labels are also consistent with the model and training data. In this work we propose IF-COMP, a scalable and efficient approximation of the pNML distribution that linearizes the model with a temperature-scaled Boltzmann influence function. IF-COMP can be used to produce well-calibrated predictions on test points as well as measure complexity in both labelled and unlabelled settings. We experimentally validate IF-COMP on uncertainty calibration, mislabel detection, and OOD detection tasks, where it consistently matches or beats strong baseline methods.

In the field of safe reinforcement learning (RL), finding a balance between satisfying safety constraints and optimizing reward performance presents a significant challenge. A key obstacle in this endeavor is the estimation of safety constraints, which is typically more difficult than estimating a reward metric due to the sparse nature of the constraint signals. To address this issue, we introduce a novel framework named Feasibility Consistent Safe Reinforcement Learning (FCSRL). This framework combines representation learning with feasibility-oriented objectives to identify and extract safety-related information from the raw state for safe RL. Leveraging self-supervised learning techniques and a more learnable safety metric, our approach enhances the policy learning and constraint estimation. Empirical evaluations across a range of vector-state and image-based tasks demonstrate that our method is capable of learning a better safety-aware embedding and achieving superior performance than previous representation learning baselines.

This paper presents a new generalization error analysis for Decentralized Stochastic Gradient Descent (D-SGD) based on algorithmic stability. The obtained results overhaul a series of recent works that suggested an increased instability due to decentralization and a detrimental impact of poorly-connected communication graphs on generalization. On the contrary, we show, for convex, strongly convex and non-convex functions, that D-SGD can always recover generalization bounds analogous to those of classical SGD, suggesting that the choice of graph does not matter. We then argue that this result is coming from a worst-case analysis, and we provide a refined optimization-dependent generalization bound for general convex functions. This new bound reveals that the choice of graph can in fact improve the worst-case bound in certain regimes, and that surprisingly, a poorly-connected graph can even be beneficial for generalization.

We consider the problem of estimating the learning rate in adaptive methods, such as AdaGrad and Adam. We propose Prodigy, an algorithm that provably estimates the distance to the solution $D$, which is needed to set the learning rate optimally. At its core, Prodigy is a modification of the D-Adaptation method for learning-rate-free learning. It improves upon the convergence rate of D-Adaptation by a factor of $\mathcal{O}(\sqrt{\log(D/d_0)})$, where $d_0$ is the initial estimate of $D$. We test Prodigy on 12 common logistic-regression benchmark datasets, VGG11 and ResNet-50 training on CIFAR10, ViT training on Imagenet, LSTM training on IWSLT14, DLRM training on Criteo dataset, VarNet on Knee MRI dataset, as well as RoBERTa and GPT transformer training on BookWiki. Our experimental results show that our approach consistently outperforms D-Adaptation and reaches test accuracy values close to that of hand-tuned Adam.

Inferring causal relationships as directed acyclic graphs (DAGs) is an important but challenging problem. Differentiable Causal Discovery (DCD) is a promising approach to this problem, framing the search as a continuous optimization. But existing DCD methods are numerically unstable, with poor performance beyond tens of variables. In this paper, we propose Stable Differentiable Causal Discovery (SDCD), a new method that improves previous DCD methods in two ways: (1) It employs an alternative constraint for acyclicity; this constraint is more stable, both theoretically and empirically, and fast to compute. (2) It uses a training procedure tailored for sparse causal graphs, which are common in real-world scenarios. We first derive SDCD and prove its stability and correctness. We then evaluate it with both observational and interventional data and in both small-scale and large-scale settings. We find that SDCD outperforms existing methods in convergence speed and accuracy, and can scale to thousands of variables.

The lifted multicut problem has diverse applications in the field of computer vision. Exact algorithms based on linear programming require an understanding of lifted multicut polytopes. Despite recent progress, two fundamental questions about these polytopes have remained open: Which lower box inequalities define facets, and which cut inequalities define facets? In this article, we answer the first question by establishing conditions that are necessary, sufficient and efficiently decidable. Toward the second question, we show that deciding facet-definingness of cut inequalities is NP-hard. This completes the analysis of canonical facets of lifted multicut polytopes.

What makes large language models (LLMs) impressive is also what makes them hard to evaluate: their diversity of uses. To evaluate these models, we must understand the purposes they will be used for. We consider a setting where these deployment decisions are made by people, and in particular, people's beliefs about where an LLM will perform well. We model such beliefs as the consequence of a human generalization function: having seen what an LLM gets right or wrong, people generalize to where else it might succeed. We collect a dataset of 19K examples of how humans make generalizations across 79 tasks from the MMLU and BIG-Bench benchmarks. We show that the human generalization function can be predicted using NLP methods: people have consistent structured ways to generalize. We then evaluate LLM alignment with the human generalization function. Our results show that -- especially for cases where the cost of mistakes is high -- more capable models (e.g. GPT-4) can do worse on the instances people choose to use them for, exactly because they are not aligned with the human generalization function.

Concentration inequalities play an essential role in the study of machine learning and high dimensional statistics. In this paper, we obtain unbounded analogues of the popular bounded difference inequality for functions of independent random variables with heavy-tailed distributions. The main results provide a general framework applicable to all heavy-tailed distributions with finite variance. To illustrate the strength of our results, we present applications to sub-exponential tails, sub-Weibull tails, and heavier polynomially decaying tails. Applied to some standard problems in statistical learning theory (vector valued concentration, Rademacher complexity, and algorithmic stability), we show that these inequalities allow an extension of existing results to heavy-tailed distributions up to finite variance.

Hyperspectral image super-resolution (HISR) aims to fuse a low-resolution hyperspectral image (LR-HSI) with a high-resolution multispectral image (HR-MSI) to obtain a high-resolution hyperspectral image (HR-HSI). Due to some existing HISR methods ignoring the significant feature difference between LR-HSI and HR-MSI, the reconstructed HR-HSI typically exhibits spectral distortion and blurring of spatial texture. To solve this issue, we propose a multi-scale feature transfer network (MFTN) for HISR. Firstly, three multi-scale feature extractors are constructed to extract features of different scales from the input images. Then, a multi-scale feature transfer module (MFTM) consisting of three improved feature matching Transformers (IMatchFormers) is designed to learn the detail features of different scales from HR-MSI by establishing the cross-model feature correlation between LR-HSI and degraded HR-MSI. Finally, a multiscale dynamic aggregation module (MDAM) containing three spectral aware aggregation modules (SAAMs) is constructed to reconstruct the final HR-HSI by gradually aggregating features of different scales. Extensive experimental results on three commonly used datasets demonstrate that the proposed model achieves better performance compared to state- of-the-art (SOTA) methods.