论文检索

Reinforcement Learning · Multi-agent

Junyu Zhang, Feihong Yang, Jian Wang, Chao Wang, Xudong Zhang

The Policy-Space Response Oracles (PSRO) framework scales equilibrium computation to large zero-sum games by iteratively expanding a restricted strategy set using deep reinforcement learning (DRL). A central challenge is to construct, under limited computational budgets, a small strategy population whose induced game well approximates the full game. Existing PSRO variants typically expand the population using best responses to meta-strategies computed from restricted-game payoffs, which can lead to inefficient expansions that provide limited global improvement. We propose to guide population expansion by directly evaluating the post-expansion population quality. Specifically, we adopt Population Exploitability (PE) to measure how well a restricted strategy set represents the full game, and introduce a two-phase exploration--selection framework that explicitly minimizes PE during expansion. We instantiate this framework as Global PSRO, a practical DRL-based algorithm that efficiently generates candidate responses and estimates PE via parameter-sharing conditional neural networks. Experiments across multiple two-player zero-sum games show that Global PSRO achieves lower exploitability and approximates Nash equilibria with significantly fewer policy iterations than prior PSRO methods.

Deep Learning · Everything Else

Mohammad Mohaiminul Islam, Rishabh Anand, David Wessels, Friso de Kruiff, Thijs Kuipers, ZHITAO YING, Clara Sánchez, Sharvaree Vadgama, Georg Bökman, Erik Bekkers

While widespread, Transformers lack inductive biases for geometric symmetries common in science and computer vision. Existing equivariant methods often sacrifice the efficiency and flexibility that make Transformers so effective through complex, computationally intensive designs. We introduce the Platonic Transformer to resolve this trade-off. By defining attention relative to reference frames from the Platonic solid symmetry groups, our method induces a principled weight-sharing scheme. This enables combined equivariance to continuous translations and Platonic symmetries, while preserving the exact architecture and computational cost of a standard Transformer. Furthermore, we show that this attention is formally equivalent to a dynamic group convolution, which reveals that the model learns adaptive geometric filters and enables a *highly scalable, linear-time convolutional variant*. Across diverse benchmarks in computer vision (CIFAR-10), 3D point clouds (ScanObjectNN), and molecular dynamics, property prediction and generation (OMol25, ProteinMD, QM9), the Platonic Transformer achieves competitive performance by leveraging these geometric constraints at no additional cost.

Theory · Online Learning and Bandits

Mingyi Li, Taira Tsuchiya, Kenji Yamanishi

This work studies online episodic tabular Markov decision processes (MDPs) with known transitions and develops best-of-both-worlds algorithms that achieve refined data-dependent regret bounds in the adversarial regime and variance-dependent regret bounds in the stochastic regime. We quantify MDP complexity using a first-order quantity and several new data-dependent measures for the adversarial regime, including a second-order quantity and a path-length measure, as well as variance-based measures for the stochastic regime. To adapt to these measures, we develop algorithms based on global optimization and policy optimization, both built on optimistic follow-the-regularized-leader with log-barrier regularization. For global optimization, our algorithms achieve first-order, second-order, and path-length regret bounds in the adversarial regime, and in the stochastic regime, they achieve a variance-aware gap-independent bound and a variance-aware gap-dependent bound that is polylogarithmic in the number of episodes. For policy optimization, our algorithms achieve the same data- and variance-dependent adaptivity, up to a factor of the episode horizon, by exploiting a new optimistic $Q$-function estimator. Finally, we establish regret lower bounds in terms of data-dependent complexity measures for the adversarial regime and a variance measure for the stochastic regime, implying that the regret upper bounds achieved by the global-optimization approach are nearly optimal.

Deep Learning · Large Language Models

Hyunjin Cho, Youngji Roh, Jaehyung Kim

As large language models are increasingly deployed in high-stakes settings, there is a growing need for tools that audit not only model outputs but also the internal computations that produce them. Circuit analysis is a central approach in mechanistic interpretability, but it is typically target-conditioned, explaining a single prompt paired with a chosen completion. This target-conditioned setup obscures mechanistic heterogeneity and hinders scalable discovery. We introduce distribution-level unsupervised feature discovery, which discovers interpretable clusters across a prompt’s continuation distribution and provides a knob to trade off semantic granularity against mechanistic specificity, without manual target selection. Our method samples continuations, represents each with (i) a semantic embedding and (ii) a mechanistic signature derived from sparse feature attributions, and clusters them via a rate–distortion objective that trades off semantic coherence and mechanistic consistency. We also show that our method has cluster-level causality, which validates the discovery of cluster-level mechanistic representation. Overall, our approach complements circuit analysis and behavioral evaluation by providing a scalable, unsupervised audit of the mechanisms underlying a model’s continuation distribution.

General Machine Learning · Methodology

Junhyoung Chung, Euijong Song, Won Hwa Kim, Gunwoong Park

We introduce Convex Distance Operator Transport (CDOT), the first convex optimal transport framework that aligns distributions across heterogeneous domains by jointly preserving feature correspondence and intrinsic geometric structure. Specifically, CDOT employs an operator-based regularization that aligns aggregated distance structures by introducing distance and conditional expectation operators. Consequently, the proposed regularization improves the robustness to local geometric variations. We further prove that the resulting CDOT discrepancy is a valid pseudometric on the space of attributed compact metric-measure spaces. In addition, we characterize the relationship between CDOT and Gromov--Wasserstien (GW) through a new notion of dispersion gap, formally elucidating the geometric source of non-convexity in GW compared to the convexity of CDOT. In the finite-sample regime, we derive a non-asymptotic risk bound decomposed into optimization and statistical errors, establishing risk consistency under a globally convergent Frank--Wolfe algorithm. Experiments on synthetic point clouds, brain connectomes, and graph classification benchmarks demonstrate better performance over existing methods, with stable and reliable behavior in practice.

General Machine Learning · Evaluation

Zhuo Chen, Oriol Mayné i Comas, Zhuotao Jin, Di Luo, Marin Soljačić

Evaluating long-context language models on natural language conflates architectural capacity to capture dependencies with semantic knowledge and vocabulary statistics. When models fail at long contexts, we cannot determine whether failures stem from fundamental architectural limitations or insufficient domain knowledge, preventing clean diagnosis of efficient architectures before expensive training on real data. We introduce **L-CUBE** (Long-Context Utilization Benchmark), a synthetic benchmark that isolates dependency-capturing capacity from semantic knowledge through hierarchical Gaussian sequences with controllable bipartite mutual information scaling. The generator provides exact ground-truth conditionals that scale efficiently to arbitrarily long sequences, enabling unconfounded evaluation via conditional KL divergence rather than perplexity alone. We define *long-context utilization* to measure the amount of available predictive information that models extract as context grows. Experiments across transformers, state space models, and efficient alternatives validate L²M capacity theory predictions and uncover new phenomena. L-CUBE enables practitioners to test whether a particular design will maintain long-context capability at target sequence lengths before committing to real-data training.

Social Aspects · Alignment

Etienne Casanova, Rafal Kocielnik, R. Michael Alvarez

Pre-trained Large Language Models (LLMs) are increasingly used for zero-shot annotation and LLM-as-a-judge tasks, yet their reliability hinges on how pre-trained priors interact with user-provided instructions. We investigate three dimensions of this interaction: (1) how an LLM’s familiarity with data and task definitions affects performance, (2) the extent to which additional information in prompts can correct zero-shot errors (“decision stickiness”), and (3) model susceptibility to misaligned task definitions. Through experiments on toxicity detection across diverse datasets (spanning social media, gaming, news, and forums) using both dense and mixture-of-experts models, we find that nearly two-thirds of zero-shot errors are resistant to correction, with an overall rescue rate (fraction of initial errors corrected by prompting) of only 36.4%. High-confidence errors prove especially resistant to correction. When given misaligned definitions, LLMs follow them while maintaining confidence levels unchanged from the aligned condition. Crucially, we introduce Definition-Specific Familiarity (DSF), a metric measuring alignment between a model’s internal concept and the task definition. After controlling for dataset-level confounds, DSF shows positive association with model performance (partial r = +0.34), while text memorization as measured by ROUGE-L shows no positive association (partial r = −0.19). Overall, these findings suggest clear limits on prompt-based correction in annotation tasks and underscore the importance of definition alignment over text-level memorization.

Deep Learning · Large Language Models

Yanzheng Xiang, Lan Wei, Yizhen Yao, Qinglin Zhu, Hanqi Yan, Chen Jin, Philip Teare, Dandan Zhang, Lin Gui, Amrutha Saseendran 等

Parallel diffusion decoding can accelerate diffusion language model inference by unmasking multiple tokens per step, but aggressive parallelism often harms quality. Revocable decoding mitigates this by rechecking earlier tokens, yet we observe that existing verification schemes frequently trigger flip-flop oscillations, where tokens are remasked and later restored unchanged. This behaviour slows inference in two ways: remasking verified positions weakens the conditioning context for parallel drafting, and repeated remask cycles consume the revision budget with little net progress. We propose COVER (Cache Override Verification for Efficient Revision), which performs leave-one-out verification and stable drafting within a single forward pass. COVER constructs two attention views via KV cache override: selected seeds are masked for verification, while their cached key value states are injected for all other queries to preserve contextual information, with a closed form diagonal correction preventing self leakage at the seed positions. COVER further prioritises seeds using a stability aware score that balances uncertainty, downstream influence, and cache drift, and it adapts the number of verified seeds per step. Across benchmarks, COVER markedly reduces unnecessary revisions and yields faster decoding while preserving output quality.

Theory · Reinforcement Learning and Planning

Orin Levy, Aviv Rosenberg, Alon Peled-Cohen, Yishay Mansour

We introduce OPO-CMDP, the first policy optimization algorithm for stochastic Contextual Markov Decision Process (CMDPs) under general offline function approximation. Our approach achieves a high probability regret bound of $\widetilde{O}(H^4\sqrt{T|S||A|\log(|\mathcal{F}||\mathcal{P}|)}),$ where $S$ and $A$ denote the state and action spaces, $H$ the horizon length, $T$ the number of episodes, and $\mathcal{F}, \mathcal{P}$ the finite function classes used to approximate the losses and dynamics, respectively. This is the first regret bound with optimal dependence on $|S|$ and $|A|$, directly improving the current state-of-the-art (Qian, Hu, and Simchi-Levi, 2024). These results demonstrate that optimistic policy optimization provides a natural, computationally superior and theoretically near-optimal path for solving CMDPs.

Theory · Online Learning and Bandits

Daniel Ezer, Alon Peled-Cohen, Yishay Mansour

We study the stochastic linear bandits with parameter noise model, in which the reward of action $a$ is $a^\top \theta$ where $\theta$ is sampled i.i.d. We show a regret upper bound of $\widetilde{O} (\sqrt{d T \log(K/\delta) \sigma^2_{\max}})$ for a horizon $T$, general action set of size $K$ of dimension $d$, and where $\sigma^2_{\max}$ is the maximal variance of the reward for any action. We further provide a lower bound of $\widetilde{\Omega} (d \sqrt{T \sigma_{\max}^2})$ which is tight (up to logarithmic factors) whenever $\log K \approx d$. For more specific action sets, $\ell_p$ unit balls with $p \leq 2$ and dual norm $q$, we show that the minimax regret is $\widetilde{\Theta} (\sqrt{dT \sigma_q^2})$, where $\sigma_q^2$ is a variance-dependent quantity that is always at most $4$. This is in contrast to the minimax regret attainable for such sets in the classic additive noise model where the regret is of order $d \sqrt{T}$. Surprisingly, we show that this optimal (up to logarithmic factors) regret bound is attainable using a very simple explore-exploit algorithm.

Reinforcement Learning · Everything Else

Oran Ridel, Alon Peled-Cohen

We study *reward-free* and *reward-agnostic* exploration in episodic finite-horizon Markov decision processes (MDPs), where an agent explores an unknown environment without observing external rewards. Reward-free exploration aims to enable $\epsilon$-optimal policies for *any* reward revealed after exploration, while reward-agnostic exploration targets $\epsilon$-optimality for rewards drawn from a small finite class. In the *reward-agnostic setting*, Li, Yan, Chen, and Fan (2024) achieve minimax sample complexity, but only for restrictively small accuracy parameter $\epsilon$. We propose a new algorithm that significantly relaxes the requirement on $\epsilon$. Our approach is novel and of technical interest by itself. Our algorithm employs an online learning procedure with carefully designed rewards to construct an exploration policy, which is used to gather data sufficient for accurate dynamics estimation and subsequent computation of an $\epsilon$-optimal policy once the reward is revealed. Finally, we establish a tight lower bound for *reward-free exploration*, closing the gap between known upper and lower bounds.

Applications · Chemistry, Physics, and Earth Sciences

Wenhan Gao, Jingxiang Qu, Yi Liu

Diffusion models typically operate in fixed-dimensional metric spaces, whereas geometric molecular data vary in dimensionality because molecules differ in size (number of atoms). A common adaptation in diffusion models for geometric molecular generation is to use architectures that handle variable-sized inputs, such as graph neural networks and transformers. **However, these approaches ignore that molecular size also sets the spatial scale of atomic coordinates, which induces inconsistent generative trajectories across sizes.** In 3D molecular diffusion, generation can be seen as forming a coarse structure first and then refining atomic positions. Larger molecules form coarse structures earlier than smaller ones because their spatial scales are larger relative to the noise. This makes the generative process inconsistent across sizes, with trajectories driven by molecular size rather than by a unified generative pattern. We are the first to identify and analyze this size-induced inconsistency by decomposing denoising dynamics, showing how spatial scale shapes formation of both 3D structure and atom types. Based on this, we propose Scaling the Prior (StP), which rescales the prior distribution by molecular size to normalize learning and generation across sizes, harmonize denoising trajectories, and enable consistently high-quality molecules.

Optimization · Discrete and Combinatorial Optimization

Zubin Zheng, Jiahao Wu, Shengcai Liu

The quantum approximate optimization algorithm (QAOA) holds promise for combinatorial optimization but is constrained by limited qubits. While divide-and-conquer frameworks like QAOA$^2$ address scalability by partitioning graphs into subgraphs, existing methods suffer from two fundamental limitations: i) misalignment between heuristic partitioning metrics and quantum optimization goals, and ii) topology-blind parameter initialization that leads to optimization cold starts. To bridge these gaps, we propose **Neural QAOA$^2$**, an end-to-end differentiable framework that jointly generates graph partitions and initial parameters. By integrating a generative evaluative network (GEN), our method utilizes a differentiable quantum evaluator as a high-fidelity performance surrogate to provide direct gradient guidance, enabling the joint generator to learn the intrinsic mapping from graph topology to high-quality partition and parameter configurations. Extensive experiments on 183 QUBO, Ising, and MaxCut instances (21 to 1000 variables) demonstrate that our gradient-driven approach significantly outperforms heuristic baselines, ranking first on 101 instances. It exhibits zero-shot generalization across out-of-distribution graph topologies and scales.

Deep Learning · Large Language Models

Guangyue Peng, Zongchao Chen, Wen Luo, Yuntao Wen, Wei Li, Ruixiang Feng, Ran Le, Chen Yang, Zhenwei An, Yang Song 等

Reverse Chain-of-Thought Generation (RCG) synthesizes reasoning traces from query-answer pairs, but runs the risk of producing post-hoc rationalizations: when models can see the answer during generation, the answer serves as a cognitive anchor that shapes the entire explanation. We formalize this phenomenon through a three-level measurement hierarchy: lexical, entropic, and probabilistic anchoring, each captures surface artifacts, entropy dynamics, and latent answer dependence, respectively. We analyze semantic suppression, the intuitive mitigation strategy that instructs models to ignore the answer, to find out its counterproduction: while it reduces lexical overlap, it paradoxically increases entropic and probabilistic anchoring. Drawing on Ironic Process Theory from cognitive psychology, we attribute this failure to active monitoring of the forbidden answer, which inadvertently deepens dependence on it. To break this cycle, we propose Structural Skeleton-guided Reasoning (SSR), a two-phase approach that first generates an answer-invariant functional skeleton structure, then uses this skeleton to guide full trace generation. By redirecting the information flow to structural planning rather than answer monitoring, SSR consistently reduces anchoring across all three levels. We further introduce Distilled SSR (SSR-D), which fine-tunes models on teacher-generated SSR traces to ensure reliable structural adherence. Experiments across open-ended reasoning benchmarks demonstrate that SSR-D achieves up to 10\% improvement over suppression baselines while preserving out-of-distribution (OOD) generalization. Code and data will be open-sourced upon acceptance.

Jianhui Chen, Yuzhang Luo, Liangming Pan

Mechanistic Interpretability has successfully identified functional circuits in Large Language Models (LLMs), yet their causal origins in the training data remain poorly understood. We bridge this gap by introducing **Mechanistic Data Attribution (MDA)**, a scalable framework that traces the formation of specific interpretable units back to training samples using Influence Functions. Through extensive pre-training experiments on the Pythia family, we causally validate that removing a small fraction of high-influence samples significantly hinders the emergence of targeted heads, whereas augmenting them accelerates formation—effects that random interventions fail to replicate. Leveraging MDA, we reveal that highly repetitive structural data—such as LaTeX and HTML—acts as a "catalyst" that significantly accelerates the emergence of induction heads. Furthermore, we observe that interventions targeting induction head formation induce a concurrent change in the model’s in-context learning (ICL) capability. This provides direct causal evidence for the long-standing hypothesis regarding the functional link between induction heads and ICL. Finally, we propose a mechanistic data augmentation pipeline that builds upon these insights to consistently accelerate mechanistic convergence across diverse model scales, offering a principled methodology for understanding and steering the fine-grained development of LLM behaviors.

Addison J. Wu, Ryan Liu, Xuechunzi Bai, Thomas Griffiths

As large language models (LLMs) are adopted into frameworks that grant them the capacity to make real decisions, it is increasingly important to ensure that they are unbiased. In this paper, we argue that the predominant approach of simply removing existing biases from models is not enough. Using a paradigm from the psychology literature, we demonstrate that LLMs can spontaneously develop novel social biases about artificial demographic groups even when no inherent differences exist. These biases result in highly stratified task allocations, which are less fair than assignments by human participants and are exacerbated by newer and larger models. In social science, emergent biases like these have been shown to result from exploration-exploitation trade-offs, where the decision-maker explores too little, allowing early observations to strongly influence impressions about entire demographic groups. To alleviate this effect, we examine a series of interventions targeting model inputs, problem structure, and explicit steering. We find that explicitly incentivizing exploration most robustly reduces stratification, highlighting the need for better multifaceted objectives to mitigate bias. These results reveal that LLMs are not merely passive mirrors of human social biases, but can actively create new ones from experience, raising urgent questions about how these systems will shape societies over time.

General Machine Learning · Evaluation

Yadong Niu, TIANZI WANG, Heinrich Dinkel, Xingwei Sun, Jiahao Zhou, Gang Li, Jizhong Liu, Xunying Liu, Jian Luan

While large audio-language models have advanced open-ended audio understanding, they still fall short of nuanced human-level comprehension. This gap persists largely because current benchmarks, limited by data annotations and evaluation metrics, fail to reliably distinguish between generic and highly detailed model outputs. To this end, this work introduces MECAT, a Multi-Expert Constructed Benchmark for Fine-Grained Audio Understanding Tasks. Generated via a pipeline that integrates analysis from specialized expert models with Chain-of-Thought large language model reasoning, MECAT provides multi-perspective, fine-grained captions and open-set question-answering pairs. The benchmark is complemented by a novel metric: DATE (Discriminative-Enhanced Audio Text Evaluation). This metric penalizes generic terms and rewards detailed descriptions by combining single-sample semantic similarity with cross-sample discriminability. A comprehensive evaluation of state-of-the-art audio models is also presented, providing new insights into their current capabilities and limitations. The data and code will be made publicly available.

Adam Karvonen, James Chua, Clément Dumas, Kit Fraser-Taliente, Subhash Kantamneni, Julian Minder, Euan Ong, Arnab Sen Sharma, Daniel Wen, Owain Evans 等

Large language model (LLM) activations are notoriously difficult to understand, with most existing techniques using complex, specialized methods for interpreting them. Recent work has proposed a simpler approach known as LatentQA: training LLMs to directly accept LLM activations as inputs and answer arbitrary questions about them in natural language. However, prior work has focused on narrow task settings for both training and evaluation. In this paper, we instead take a generalist perspective. We evaluate LatentQA-trained models, which we call Activation Oracles (AOs), in far out-of-distribution settings and examine how performance scales with training data diversity. We find that AOs can recover information fine-tuned into a model (e.g., biographical knowledge or malign propensities) that does not appear in the input text, despite never being trained with activations from a fine-tuned model. Our main evaluations are four downstream tasks where we can compare to prior white- and black-box techniques. We find that even narrowly-trained LatentQA models can generalize well, and that adding additional training datasets (such as classification tasks and a self-supervised context prediction task) yields consistent further improvements. Our best AOs match or exceed white-box baselines on all four tasks and the best overall baseline on 3 of 4. These results suggest that diversified training to answer natural-language queries imparts a general capability to verbalize information about LLM activations.

Applications · Robotics

John Won, Kyungmin Lee, Huiwon Jang, Dongyoung Kim, Jinwoo Shin

Augmenting Vision-Language-Action models (VLAs) with world models is promising for robotic policy learning but faces challenges in jointly predicting states and actions due to the modality gap. To address this, we propose DUal-STream diffusion (DUST), a world-model augmented VLA framework featuring a multimodal diffusion transformer that maintains separate modality streams while enabling cross-modal knowledge sharing. In addition, DUST utilizes independent noise perturbations and a decoupled flow matching loss to learn cross-modal causal relationships. We further introduce an asynchronous sampling method for action and vision tokens that enhances performance through inference-time scaling. Experimental results on simulated benchmarks like RoboCasa and GR-1 show that DUST achieves up to 6\% gains over state-of-the-art VLA and world-modeling baselines, with inference-time scaling providing an additional 2–5\% improvement. In real-world tasks using the Franka Research 3, DUST outperforms baselines by 10\% in success rate. Finally, we demonstrate that DUST enables effective transfer learning through both pretraining on action-free videos and joint-training with heterogeneous robot and human datasets.

José Ramón Enríquez, Jiaxin Pei, Alex Pentland

AI systems can strengthen democracy by supporting deliberation at scale by addressing cognitive, social, platform-design, and market-driven frictions, while preserving human agency. Unlike proposals such as liquid democracy that restructure representation through vote delegation, in this position paper, we argue that AI-assisted deliberation offers a more promising path by lowering barriers to meaningful engagement without substituting machine judgment for human choice. Drawing on evidence from online platforms and experimental research, we identify four guiding principles: preserving agency and autonomy, encouraging mutual respect, promoting equality and inclusiveness, and augmenting rather than substituting active citizenship. We also address critical challenges, including alignment, sycophancy, training bias, and over-reliance on AI systems. We call on the machine learning community to develop deliberation-focused AI systems evaluated not on engagement metrics but on their capacity to facilitate informed, representative, and friction-robust discourse.