ICML 2026 论文列表

Multi-agent LLM debates are increasingly used in domains such as policy, politics, and city planning, where ground truth is often unavailable. Yet existing evaluations rely heavily on outcome-based proxies such as consensus, majority vote, or LLM-as-judge scores, which can miss failures like sycophancy, domination, and premature convergence. We introduce a diagnostic framework that evaluates both debate outcomes and the deliberative process using interpretable metrics for engagement, responsiveness, influence asymmetry, balance, stability, and agent utility. Across real-world debate settings and validation benchmarks, our process-level diagnostics align more closely with human judgments and reveal interaction failures that standard outcome-only measures overlook. These results show that reliable evaluation of multi-agent debates requires measuring not only what answer agents reach, but how they reach it.

Ensuring fair and equitable treatment across diverse groups, particularly in multi-class classification tasks, poses a significant challenge due to the persistent biases inherent in machine learning models. Most existing bias mitigation techniques are tailored to binary settings, and the presence of multi-dimensional outputs and complex fairness mechanisms makes their extension to multi-class scenarios neither straightforward nor effective. In this paper, we investigate two fundamental, unresolved challenges in fair classification: (i) *characterizing the optimal accuracy-fairness frontier in multi-class settings*, and (ii) *designing practical algorithms that attain this optimum across different training phases*. To tackle these challenges, we first specify an analytically tractable probabilistic formulation of the optimal classifier under fairness constraints. Building upon this, we propose two attribute-blind algorithms to enforce fairness requirements in practice: an in-processing approach for fairness intervention during training via the reduction approach, and a post-processing approach for fine-tuning output probabilities with plug-in estimation. Theoretical analysis reveals that both methods converge to the optimal accuracy-fairness Pareto frontier. Experiments conducted on multiple datasets demonstrate the superior performance of our methods in balancing accuracy and fairness.

Multi-token prediction (MTP) is a prominent strategy to significantly speed up generation in large language models (LLMs), especially in byte-level LLMs, which are tokeniser-free but prohibitively slow. However, many existing MTP methods either assume independence between future tokens, sacrificing expressiveness, or generate tokens one at a time within the window, increasing latency. In this work, we investigate the trade-off between expressiveness and latency in MTP within the framework of probabilistic circuits (PCs). Our framework, MTPC, allows one to explore different ways to encode the joint distributions over future tokens by selecting circuit architectures, generalising classical models such as (hierarchical) mixture models, hidden Markov models, and tensor networks. We show the efficacy of MTPC by retrofitting existing byte-level LLMs, such as EvaByte, and byte-fied subword models, such as Llama3.2 3B. Our experiments show that, when combined with speculative decoding, MTPC substantially speeds up generation compared to MTP with independence assumptions, while guaranteeing to retain the performance of the original verifier LLM. We also rigorously study the optimal trade-off between expressiveness and latency when exploring the possible parameterisations of MTPC, such as PC architectures and partial layer sharing between the verifier and draft LLMs.

Agents powered by advanced large language models (LLMs) have demonstrated impressive capabilities across diverse complex applications. Recently, Multi-Agent Systems (MAS), wherein multiple agents collaborate and communicate with each other, have exhibited enhanced capabilities in complex tasks, such as high-quality code generation and arithmetic reasoning. However, the development of such systems often relies on handcrafted methods, and the literature on systematic design and optimization of LLM-based MAS remains limited. In this work, we introduce OMAC, a general framework designed for holistic optimization of LLM-based MAS. Specifically, we identify five key optimization dimensions for MAS, encompassing both agent functionality and collaboration structure. Building upon these dimensions, we first propose a general algorithm, utilizing two actors termed the Semantic Initializer and the Contrastive Comparator, to optimize any single dimension. Then, we present an algorithm for joint optimization across multiple dimensions. Extensive experiments demonstrate the superior performance of OMAC on diverse tasks against recent approaches. Codes are available at: https://github.com/xiwenchao/OMAC.

The anchor-based multi-view clustering methods have received extensive attention due to their efficiency and scalability in large-scale data scenarios. Existing anchor-based methods still face challenges in learning compact and semantically discriminative anchors. Current mainstream approaches typically rely on random sampling strategies or orthogonal constraints for anchor selection and learning. However, they often optimize anchor learning and cluster assignment in a relatively separate manner, leaving the clustering semantics in the sample space insufficiently exploited for calibrating the anchor space. As a result, the learned anchors may suffer from redundant coverage and ambiguous cluster boundaries. Unlike existing anchor-based multi-view clustering methods, this paper proposes a Dual-Calibration Multi-view Clustering via Compact Anchor Learning (DCMC), which effectively improves anchor quality through a dual-space alignment mechanism. Specifically, DCMC initializes view-specific anchors to capture the underlying data distribution, and then enforces bidirectional consistency between the anchor space and the clustering space to jointly optimize both the sample-to-anchor assignments and the cluster assignments. The alternating optimization process effectively enhances cross-view semantic consistency while preserving the discriminative characteristics of each view. Experimental results demonstrate that DCMC outperforms state-of-the-art methods across multiple benchmark tests, confirming its effectiveness and reliability.

Recent advances in large language models (LLMs) have substantially expanded the scope of Table Question Answering (TableQA). However, existing benchmarks primarily treat TableQA as a passive, single-turn natural language understanding task, lacking the capacity to evaluate autonomous reasoning and tool-call trajectories in realistic, multi-turn scenarios. To bridge this gap, we introduce TableAgent-Bench, a large-scale bilingual benchmark that reformulates TableQA as proactive, agentic interactions over structurally complex, multi-table environments. With a topology-aware construction strategy, TableAgent-Bench captures dynamic intent evolution through 1,310 multi-turn dialogues grounded in 2,275 industrial tables. Furthermore, we propose the Table-centric Agent Evaluation Framework (TAEF) to assess agent interactions with complex table structures. Specifically, TAEF integrates a specialized agent toolset and 4 metric categories to systematically diagnose intermediate failure modes, assessing performance across table localization, tool-invocation rationality, and trajectory-level pass rate. Extensive experiments with 25 state-of-the-art LLM agents reveal a substantial capability gap, with even the strongest model Gemini-3-Pro-Preview achieving only 53.4% information coverage. We expect TableAgent-Bench to serve as a rigorous testbed for developing and evaluating agents capable of robust table-centric reasoning.

Extracting skills from multi-agent offline dataset improves learning efficiency via sharing task-invariant coordination skills among tasks. In settings where tasks occur sequentially and the space of skills grows exponentially, existing approaches that rely on heuristically designed and fixed-sized skill libraries struggle to resolve the problem of distributional shift and interference, facing catastrophic forgetting and plasticity loss. To address this problem and endow agents with the ability to continually discover and reuse coordination skills in open-environment, we propose COMAD, a principled framework for **C**ontinual **O**ffline **M**ulti-**a**gent Skill **D**iscovery via Skill Partition and Reuse. We first discover skills from mixed multi-agent behavior data with an auto-encoder to transform coordination knowledge into reusable coordination skills. Then we construct a skill-augmented policy learning objective with multi-head architectures, explicitly guiding the advantage function with reusable skills identified via a density-based reusability estimator. Theoretical analysis shows our method approximates the optimum of a continual skill discovery problem. Empirical results across diverse MARL benchmarks show that COMAD continually expands its skill library to mitigate interference, achieving superior forward and backward transfer for task streams compared to multiple baselines.

We examine the interplay between ordinal, preference-based solution concepts in games and the outcomes of payoff-driven learning dynamics, asking to what extent the combinatorial data of a game—its *preference graph*—can predict the long-run behavior of no-regret dynamics such as *follow-the-regularized-leader* (FTRL). In one direction, we show that the skeleton of every *dynamically stable* set (i.e., the set of pure profiles it contains) must also be *preferentially stable*, that is, it must be closed under profitable deviations. We then ask the converse question: when are preferences sufficient to describe the long-run behavior of the players' learning dynamics? We begin by showing that preferences are indeed enough to fully characterize asymptotic stability in the case of *subgames*—i.e., subsets of pure profiles obtained by restricting players' action sets. Beyond this case however, the equivalence between dynamic and preferential stability breaks down: in particular, we construct a three-player game with a preferentially stable set whose span is dynamically *unstable*, showing that preferences are *not sufficient* to describe dynamically stable behavior in general. To restore stability, we introduce the notion of *leaklessness*, a measure of aggregate payoff drift away from a set of pure profiles, and we use it to identify a payoff-based condition guaranteeing that the span of a set of pure profiles is stable and attracting.

The general-purpose nature of Large Language Models (LLMs) presents a significant challenge for domain-specific applications, often leading to out-of-domain (OOD) interactions that undermine the provider's intent. Existing methods for detecting such scenarios treat the LLM as an uninterpretable black box and overlook the internal processing of inputs. In this work we show that layer transitions provide a promising avenue for extracting domain-specific signature. Specifically, we present several lightweight ways of learning on internal dynamics encoded using a sparse autoencoder (SAE) that exhibit great capability in distinguishing OOD texts. Building on top of SAEs representation transitions enables us to better interpret the LLM internal evolution of input processing and shed light on its decisions. We provide a comprehensive analysis of the method and benchmark it with the gemma-2 2B and 9B models. Our results emphasize the efficacy of the internal process in capturing fine-grained input-related details.

Link prediction (LP) is a fundamental problem in graph learning and can be broadly categorized into node-based and subgraph-based approaches. While subgraph-based LP methods often achieve superior predictive performance by exploiting localized structural information, they suffer from efficiency bottlenecks due to the high computational cost of per-query subgraph message-passing during both training and inference. To address this challenge, we propose MIMO-LP, a Multi-Input Multi-Output (MIMO) framework that accelerates subgraph-based LP. Given a batch of query node pairs and their corresponding subgraphs extracted from a shared full graph, MIMO-LP superposes their message-passing processes into a shared latent space while ensuring minimal interference among them. This design enables MIMO-LP to multiplex multiple queries within a single forward pass during both training and inference, substantially reducing redundant message-passing computations in overlapping subgraph regions. Extensive experiments demonstrate that MIMO-LP achieves a 14x-44x speedup over existing one-to-one subgraph-based methods, while maintaining comparable predictive performance.

Multi-agent LLM debates achieve strong performance on decision-making tasks as well as problem-solving benchmarks, yet their safety and fairness risks remain poorly understood. Notably, interaction can amplify the biases of single LLMs, raising concerns for real-world deployment. We identify the emergence of collective (often biased) norms in multi-agent LLM debates and show that noise (e.g., LLM sampling temperature) is a key driver. To explain this, we propose an analytical framework drawing on physics-inspired theoretical models of social dynamics. We predict a phase transition to collective bias when conformity surpasses a critical threshold given the LLMs' initial bias and debate noise. We test the theoretical predictions through controlled experiments and observe a finite-size crossover consistent with an underlying phase transition. We further find that agent heterogeneity suppresses emergence by smoothing (rounding) this transition. Finally, we show that these insights generalize to realistic decision-making tasks, including investment decisions and LLM-as-a-judge evaluation.

LLM-based multi-agent systems excel at planning, tool use, and role coordination, but their openness and interaction complexity also expose them to jailbreak and adversarial collaboration. Existing defenses fall into two lines: (i) self-verification that asks each agent to pre-filter unsafe instructions before execution, and (ii) external guard modules that police behaviors. The former often underperforms because a standalone agent lacks sufficient capacity to detect cross-agent unsafe chains and delegation-induced risks; the latter increases system overhead and creates a single-point-of-failure—once compromised, system-wide safety collapses, and adding more guards worsens cost and complexity. To solve these challenges, we propose AdvEvo-MARL, a co-evolutionary multi-agent reinforcement learning framework that internalizes safety into task agents. Rather than relying on external guards, AdvEvo-MARL jointly optimizes attackers (which synthesize evolving jailbreak prompts) and defenders (task agents trained to both accomplish their duties and resist attacks) in adversarial learning environments. To stabilize learning and foster cooperation, we introduce a public baseline for advantage estimation: agents within the same functional group share a group-level mean-return baseline, enabling lower-variance updates and stronger intra-group coordination. Across representative attack scenarios, AdvEvo-MARL consistently keeps attack-success rate (ASR) below 20\%, whereas baselines reach up to 38.33\%, while preserving or even improving task accuracy (up to +3.67\%). These results show that safety and utility can be jointly improved without relying on extra guard agents or added system overhead.

Reinforcement learning usually assumes a given or sometimes even fixed environment in which an agent seeks an optimal policy to maximize its long-term discounted reward. In contrast, we consider agents that are not limited to passive adaptations: they instead have model-changing actions that actively modify the RL model of world dynamics itself. Reconfiguring the underlying transition processes can potentially increase the agents' rewards. Motivated by this setting, we introduce the multi-layer configurable time-varying Markov decision process (MCTVMDP). In an MCTVMDP, the lower-level MDP has a non-stationary transition function that is configurable through upper-level model-changing actions. The agent's objective consists of two parts: Optimize the configuration policies in the upper-level MDP and optimize the primitive action policies in the lower-level MDP to jointly improve its expected long-term reward.

Multi-agent reinforcement learning (MARL) has received increasing attention for solving complex decision-making tasks. Networked MARL approaches offer a decentralized solution for parameter sharing to accelerate training via value aggregation. However, existing federated aggregations rely on convex averaging that may fail to converge to global optima and cause learning rollback in the online learning setting. In this paper, we formally characterize the learning rollback phenomenon arising from aggregating value estimates with unequal uncertainty under heterogeneous online update depths. We propose a novel adaptive global consensus (AGC) mechanism for Q-value aggregation in decentralized MARL policy evaluation, which dynamically adjusts aggregation weights based on agents’ uncertainty. We establish theoretical guarantees on accelerated convergence and bounded learning variance with empirical validations, advancing the state-of-art MARL theory.

Training generalist agents capable of adapting to diverse scenarios requires interactive environments for self-exploration. However, interactive environments remain critically scarce, and existing synthesis methods suffer from significant limitations regarding environmental diversity and scalability. To address these challenges, we introduce ScaleEnv, a framework that constructs fully interactive environments and verifiable tasks entirely from scratch. Specifically, ScaleEnv ensures environment reliability through procedural testing, and guarantees task completeness and solvability via tool dependency graph expansion and executable action verification. By enabling agents to learn through exploration within ScaleEnv, we demonstrate significant performance improvements on unseen, multi-turn tool-use benchmarks such as $\tau^2$-Bench and VitaBench, highlighting strong generalization capabilities. Furthermore, we investigate the relationship between increasing number of domains and model generalization performance, providing empirical evidence that scaling environmental diversity is critical for robust agent learning.

Existing deep learning models have demonstrated potential in Diabetic retinopathy (DR) diagnosis, but they still suffer from three key challenges: reliance on single-source inputs, opaque and untraceable reasoning processes, and the absence of a mechanism for result verification. Thus, we propose a provenance-enabled concept-based framework for multi-view DR diagnostic (ProConMV), which integrates DR lesion masks, clinical text and multi-view data, utilizing multimodal prompt analysis and visual-text concept interaction to learn the interpretable multi-source input. During the reasoning stage, the proposed framework introduces lesion concepts for causal reasoning chains combining clinical guidelines, and adds doctor intervention for human-machine collaboration. For dynamic fusion decision and verification in multi-view DR diagnosis, we derive via generalization theory that incorporating each view’s lesion concept uncertainty and grading uncertainty reduces the generalization error upper bound. Accordingly, we design a dual uncertainty-aware module to enable provenance-based verification, ultimately enabling verifiable analysis of DR diagnostic results. Extensive experiments conducted on two public multi-view DR datasets demonstrate the effectiveness of our method. The code will be released at https://github.com/SoY0ung/ProConMV.

Tool-calling is a central component of modern large language model (LLM) agents, equipping them with skills beyond their parametric knowledge. This paper studies tool-calling along two complementary axes: **effectiveness**, i.e., how this capability is *measured*, and **efficiency**, i.e., how it is *learned*. On effectiveness, we systematically analyze tool-calling evaluation pipelines and show that results can be highly sensitive to seemingly minor, often undocumented implementation choices including the *random seed*, *system prompt*, *multi-turn template construction*, and how *prior interaction/reasoning history* is carried forward. These choices can lead to substantial differences in reported performance, especially in multi-turn settings where without rigorous standardization, leaderboard rankings are unreliable. On efficiency, we examine standard reinforcement learning (RL) for tool-calling and identify two sources of computational waste: (i) during rollouts, many prompts produce no learning signal, and (ii) during policy updates, optimization incurs high computational cost. Guided by these findings, we introduce two techniques that accelerate RL-based tool-calling training, achieving substantial wall-clock speedup without degrading performance.

Existing Meta-Black-Box Optimization (MetaBBO) methods focus on how to search when controlling optimizers, but largely overlook where to search. We propose MetaSG-SAEA, a bi-level MetaBBO framework for expensive constrained multi-objective optimization problems (ECMOPs), in which a meta-policy provides search guidance to the low-level Surrogate-Assisted Evolutionary Algorithm (SAEA). To achieve this, we introduce Max–Min Constraint-Calibrated Inequality (MM-CCI), a compact, problem-agnostic region abstraction that maps heterogeneous constraint evaluations to an ordered scalar level; we further provide a theoretical analysis of its fundamental properties. Building on this region abstraction, we adopt diffusion-based population initialization to translate the meta-policy’s region-level guidance into solution-level priors for the SAEA. To make MetaSG-SAEA scalable, we construct an attention-based state representation across varying problem dimensions, population sizes, and numbers of objectives and constraints. Experimental results demonstrate that MetaSG-SAEA outperforms state-of-the-art baselines across diverse benchmarks and exhibits the ability to generalize across problem distributions.

Constitutional AI has focused on single-model alignment using fixed principles. However, multi-agent systems create novel alignment challenges through emergent social dynamics. We present Constitutional Evolution, a framework for automatically discovering behavioral norms in multi-agent LLM systems. Using a grid-world simulation with survival pressure, we study the tension between individual and collective welfare, quantified via a Societal Stability Score $\mathcal{S} \in [0,1]$ that combines productivity, survival, and conflict metrics. Adversarial constitutions lead to societal collapse ($\mathcal{S}=0$), while vague prosocial principles (''be helpful, harmless, honest'') produce inconsistent coordination ($\mathcal{S}=0.249$). Even constitutions designed by Claude 4.5 Opus with explicit knowledge of the objective achieve only moderate performance ($\mathcal{S}=0.332$). Using LLM-driven genetic programming with multi-island evolution, we evolve constitutions maximizing social welfare without explicit guidance toward cooperation. The evolved constitution $\mathcal{C}^*$ achieves $\mathcal{S}=0.556\pm0.008$ (123\% higher than human-designed baselines, $N=10$), eliminates conflict, and discovers that minimizing communication (0.9\% vs 62.2\% social actions) outperforms verbose coordination. Our interpretable rules demonstrate that cooperative norms can be discovered rather than prescribed.