arXiv cs.AI· Keru Chen, Sen Lin, Yingbin Liang, Nathaniel D. Bastian, Shaofeng Zou·· 3 天前AI 评分27
CoSIR：用可验证信息奖励训练 LLM 多智能体协作
Training LLMs for Multi-Agent Collaboration with Verifiable Information Rewards
AI 导读
研究者提出 CoSIR（Collaboration via Surprisal-based Information Rewards），一种后训练方法，通过可验证环境中的贝叶斯评估器，按团队 surprisal 的降低幅度为共享信息和请求打分，并奖励智能体在自身决策中使用队友信息。
正文 · 原文
Abstract:Effective collaboration among large language model (LLM) agents in multi-agent systems requires sharing relevant evidence, requesting missing information, and incorporating teammates' contributions into decisions. Training these behaviors is challenging because final task outcomes provide limited feedback on the value of individual messages. We introduce CoSIR (Collaboration via Surprisal-based Information Rewards), a post-training approach that uses a simple, verifiable environment to teach collaboration behaviors that generalize to more complex, unseen tasks. We propose a Bayesian evaluator that scores shared information by its reduction in team surprisal, defined as the sum of the negative log probabilities assigned to the correct option given each agent's available evidence. Requests are scored by the potential reduction in the same quantity if the requested information were truthfully provided. Separate rewards encourage agents to use the received information in their private choices and calculations. We post-train Qwen3-4B, Qwen3-8B, and Gemma-3-4B-it through self-play and evaluate transfer to seven unseen natural-language collaboration tasks. CoSIR achieves the highest aggregate score, while its performance in general tasks remains competitive. Behavioral analyses show increased disclosure of private evidence and fewer judged failures to use teammates' input.
| Comments: | 23 pages |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.17893 [cs.AI] |
| (or arXiv:2610.17893v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.17893 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Keru Chen [view email]
[v1]
Tue, 13 Oct 2026 07:46:48 UTC (278 KB)
来源：arXiv cs.AI · arxiv.org