Huiming Zhang
Publications
Adaptive Robust Estimator for Multi-Agent Reinforcement Learning
Multi-agent collaboration has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models, yet it suffers from interaction-level ambiguity that blurs generation, critique, and revision, making credit assignment across agents difficult. Moreover, policy optimization in this setting is vulnerable to heavy-tailed and noisy rewards, which can bias advantage estimation and trigger unstable or even divergent training. To address both issues, we propose a robust multi-agent reinforcement learning framework for collaborative reasoning, consisting of two components: Dual-Agent Answer-Critique-Rewrite (DACR) and an Adaptive Robust Estimator (ARE). DACR decomposes reasoning into a structured three-stage pipeline: answer, critique, and rewrite, while enabling explicit attribution of each agent's marginal contribution to its partner's performance. ARE provides robust estimation of batch experience means during multi-agent policy optimization. Across mathematical reasoning and embodied intelligence benchmarks, even under noisy rewards, our method consistently outperforms the baseline in both homogeneous and heterogeneous settings. These results indicate stronger robustness to reward noise and more stable training dynamics, effectively preventing optimization failures caused by noisy reward signals.
Counterfactual Credit Policy Optimization for Multi-Agent Collaboration
Collaborative multi-agent large language models (LLMs) can solve complex reasoning tasks by decomposing roles and aggregating diverse hypotheses. Yet, reinforcement learning (RL) for such systems is often undermined by credit assignment: a shared global reward obscures individual contributions, inflating update variance and encouraging free-riding. We introduce Counterfactual Credit Policy Optimization (CCPO), a framework that assigns agent-specific learning signals by estimating each agent's marginal contribution through counterfactual trajectories. CCPO builds dynamic counterfactual baselines that simulate outcomes with an agent's contribution removed, yielding role-sensitive advantages for policy optimization. To further improve stability under heterogeneous tasks and data distributions, we propose a global-history-aware normalization scheme that calibrates advantages using global rollout statistics. We evaluate CCPO on two collaboration topologies: a sequential Think--Reason dyad and multi-agent voting. Across mathematical and logical reasoning benchmarks, CCPO mitigates free-riding and outperforms strong multi-agent RL baselines, yielding finer-grained and more effective credit assignment for collaborative LLM training. Our code is available at https://github.com/bhai114/ccpo.
Sharper Generalization Bounds for Transformer
This paper studies generalization error bounds for Transformer models. Based on the offset Rademacher complexity, we derive sharper generalization bounds for different Transformer architectures, including single-layer single-head, single-layer multi-head, and multi-layer Transformers. We first express the excess risk of Transformers in terms of the offset Rademacher complexity. By exploiting its connection with the empirical covering numbers of the corresponding hypothesis spaces, we obtain excess risk bounds that achieve optimal convergence rates up to constant factors. We then derive refined excess risk bounds by upper bounding the covering numbers of Transformer hypothesis spaces using matrix ranks and matrix norms, leading to precise, architecture-dependent generalization bounds. Finally, we relax the boundedness assumption on feature mappings and extend our theoretical results to settings with unbounded (sub-Gaussian) features and heavy-tailed distributions.