Research

Paper

AI LLM March 23, 2026

Counterfactual Credit Policy Optimization for Multi-Agent Collaboration

Authors

Zhongyi Li, Wan Tian, Yikun Ban, Jinju Chen, Huiming Zhang, Yang Liu, Fuzhen Zhuang

Abstract

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.

Metadata

arXiv ID: 2603.21563
Provider: ARXIV
Primary Category: cs.AI
Published: 2026-03-23
Fetched: 2026-03-24 06:02

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.21563v1</id>\n    <title>Counterfactual Credit Policy Optimization for Multi-Agent Collaboration</title>\n    <updated>2026-03-23T04:35:02Z</updated>\n    <link href='https://arxiv.org/abs/2603.21563v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.21563v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>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.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-03-23T04:35:02Z</published>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>Zhongyi Li</name>\n    </author>\n    <author>\n      <name>Wan Tian</name>\n    </author>\n    <author>\n      <name>Yikun Ban</name>\n    </author>\n    <author>\n      <name>Jinju Chen</name>\n    </author>\n    <author>\n      <name>Huiming Zhang</name>\n    </author>\n    <author>\n      <name>Yang Liu</name>\n    </author>\n    <author>\n      <name>Fuzhen Zhuang</name>\n    </author>\n  </entry>"
}