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 2602.03794cs.AIcs.LG
收藏

通过多样性理解基于大语言模型的多智能体系统中的智能体扩展

Understanding Agent Scaling in LLM-Based Multi-Agent Systems via Diversity

Shanghai Jiao Tong University(上海交通大学);California Institute of Technology(加州理工学院);Johns Hopkins University(约翰霍普金斯大学)

Yingxuan Yang, Chengrui Qu, Muning Wen, Laixi Shi, Ying Wen, Weinan Zhang, Adam Wierman, Shangding Gu

AI总结:

本文通过引入K*指标,揭示了基于大语言模型的多智能体系统中,异质性智能体配置比同质配置更有效,且性能受任务不确定性限制。

摘要:

基于大语言模型的多智能体系统(MAS)已出现为解决复杂任务的有前途的方法,这些任务对单个LLM来说很困难。一种自然的策略是通过增加智能体数量来提高性能;然而,我们发现这种扩展在同质设置中表现出强烈的边际收益递减,而引入异质性(例如不同模型、提示或工具)仍能产生显著收益。这提出了一个根本问题:什么限制了扩展,为什么多样性有助于提升性能?我们提出一个信息论框架,表明MAS的性能受任务固有不确定性限制,而不是智能体数量。我们推导出架构无关的界限,证明改进取决于系统访问的有效通道数量。同质智能体早期饱和,因为其输出高度相关,而异质智能体贡献互补证据。我们进一步引入K*,一个有效通道数,它量化了系统访问的有效通道数,而无需地面真实标签。实证上,我们显示异质配置始终优于同质扩展:2个多样化的智能体可以匹配或超过16个同质智能体的性能。我们的结果为通过多样性设计构建高效且稳健的MAS提供了原则性指导。代码和数据集可在链接中获得:https://github.com/SafeRL-Lab/Agent-Scaling。

英文摘要:

LLM-based multi-agent systems (MAS) have emerged as a promising approach to tackle complex tasks that are difficult for individual LLMs. A natural strategy is to scale performance by increasing the number of agents; however, we find that such scaling exhibits strong diminishing returns in homogeneous settings, while introducing heterogeneity (e.g., different models, prompts, or tools) continues to yield substantial gains. This raises a fundamental question: what limits scaling, and why does diversity help? We present an information-theoretic framework showing that MAS performance is bounded by the intrinsic task uncertainty, not by agent count. We derive architecture-agnostic bounds demonstrating that improvements depend on how many effective channels the system accesses. Homogeneous agents saturate early because their outputs are strongly correlated, whereas heterogeneous agents contribute complementary evidence. We further introduce $K^*$, an effective channel count that quantifies the number of effective channels without ground-truth labels. Empirically, we show that heterogeneous configurations consistently outperform homogeneous scaling: 2 diverse agents can match or exceed the performance of 16 homogeneous agents. Our results provide principled guidelines for building efficient and robust MAS through diversity-aware design. Code and Dataset are available at the link: https://github.com/SafeRL-Lab/Agent-Scaling.

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。

↑