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 2512.14079cs.AIcs.CLcs.MA
收藏

多智能体系统的语法搜索

Grammar Search for Multi-Agent Systems

University of Arizona(亚利桑那大学);ServiceNow(ServiceNow公司);Mila - Quebec AI Institute(魁北克AI研究院)

Mayank Singh, Vikas Yadav, Shiva Krishna Reddy Malay, Shravan Nayak, Sai Rajeswar, Sathwik Tejaswi Madhusudhan, Eduardo Blanco

AI总结:

本研究提出了一种结构化框架,通过固定组件生成多智能体系统,在数学和问答领域优于现有方法,且更经济高效。

摘要:

多智能体系统的自动搜索近期已成为代理AI研究中的关键重点。先前的几种方法依赖于基于大语言模型的自由形式搜索代码空间。在本工作中,我们提出了一种更加结构化的框架,通过一组固定的简单、可组合的组件来探索相同的空间。我们展示了,尽管在候选生成阶段缺乏大语言模型的生成灵活性,我们的方法在数学和问答两个领域中的五个基准测试中,有四个超过了先前的方法。此外,我们的方法还提供了额外的优势,包括更经济的搜索过程,以及生成具有更简单逻辑的模块化、可解释的多智能体系统。

英文摘要:

Automatic search for Multi-Agent Systems has recently emerged as a key focus in agentic AI research. Several prior approaches have relied on LLM-based free-form search over the code space. In this work, we propose a more structured framework that explores the same space through a fixed set of simple, composable components. We show that, despite lacking the generative flexibility of LLMs during the candidate generation stage, our method outperforms prior approaches on four out of five benchmarks across two domains: mathematics and question answering. Furthermore, our method offers additional advantages, including a more cost-efficient search process and the generation of modular, interpretable multi-agent systems with simpler logic.

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。

↑