arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.04234cs.MAcs.AI
收藏

何时多智能体协作有助于?从熵的角度来看

When Does Multi-Agent Collaboration Help? An Entropy Perspective

Yantai Research Institute of Harbin Engineering University(哈尔滨工程大学烟台研究室);The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州))

Yuxuan Zhao, Sijia Chen, Ningxin Su

AI总结:

本文从熵的角度探讨了多智能体协作的有效性,通过分析不同拓扑结构、六个推理基准和两个智能体任务中的熵转换,发现单个智能体在43.3%的情况下表现更优,并揭示了熵动态在第一轮交互中的决定性作用。研究提出了三个关键观察:确定性偏好、基础熵和任务意识,并引入了熵判别算法来提升智能体系统的性能。

摘要:

多智能体系统(MAS)已逐渐成为利用大型语言模型(LLMs)解决复杂任务的主流范式。然而,基于公开可用LLMs构建的MAS的有效性机制,尤其是其成功或失败的内在原因,仍鲜有研究。本文从熵的角度重新审视MAS,通过研究问题解决过程中不同拓扑结构、六个推理基准和两个智能体任务中的熵转换,考虑了智能体内部和交互动态。通过分析245个跨越token级、agent级和轮次级的熵特征,我们意外发现,在约43.3%的情况下,单个智能体优于MAS。此外,我们发现熵动态主要在首次交互轮次中决定。我们提供了三个关键观察:1)确定性偏好:峰值熵直接损害MAS的正确性,稳定熵直接促进MAS的正确性;2)基础熵:基础模型在问题解决过程中具有较低熵会因果驱动MAS性能;3)任务意识:MAS的熵动态在不同任务中扮演不同角色。基于这些见解,我们引入了一个简单而有效的算法,即熵判别器,用于从MAS的pass@k结果中选择解决方案,从而在所有MAS配置和任务中均实现了稳定的准确率提升。我们的源代码可在https://github.com/AgenticFinLab/multiagent-entropy获取。

英文摘要:

Multi-agent systems (MAS) have emerged as a prominent paradigm for leveraging large language models (LLMs) to tackle complex tasks. However, the mechanisms governing the effectiveness of MAS built upon publicly available LLMs, specifically the underlying rationales for their success or failure, remain largely unexplored. In this paper, we revisit MAS through the perspective of \textit{entropy}, considering both intra- and inter-agent dynamics by investigating entropy transitions during problem-solving across various topologies, six reasoning benchmarks, and two agentic tasks. By analyzing 245 features spanning token-, agent-, and round-level entropy, we counterintuitively find that a single agent outperforms MAS in approximately 43.3\% of cases, and that entropy dynamics are largely determined during the first round of interaction. Furthermore, we provide three key observations: 1) \textit{Certainty Preference}: peak entropy directly harms and stable entropy directly benefits MAS correctness; 2) \textit{Base Entropy}: base models with lower entropy during problem-solving causally drive MAS performance; and 3) \textit{Task Awareness}: entropy dynamics of MAS play varying roles across different tasks. Building on these insights, we introduce a simple yet effective algorithm, the \textit{Entropy Judger}, to select solutions from MAS's pass@$k$ results, leading to consistent accuracy improvements across all MAS configurations and tasks. Our source code is available at \href{https://github.com/AgenticFinLab/multiagent-entropy}{this https URL}.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑