Multi-agent Architecture Search via Agentic Supernet

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Zhang, Guibin, Niu, Luyang, Fang, Junfeng, Wang, Kun, Bai, Lei, Wang, Xiang
Format:	Preprint
Published:	2025
Subjects:	Machine Learning Computation and Language Multiagent Systems
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

_version_	1866908398334771200
author	Zhang, Guibin Niu, Luyang Fang, Junfeng Wang, Kun Bai, Lei Wang, Xiang
author_facet	Zhang, Guibin Niu, Luyang Fang, Junfeng Wang, Kun Bai, Lei Wang, Xiang
contents	Large Language Model (LLM)-empowered multi-agent systems extend the cognitive boundaries of individual agents through disciplined collaboration and interaction, while constructing these systems often requires labor-intensive manual designs. Despite the availability of methods to automate the design of agentic workflows, they typically seek to identify a static, complex, one-size-fits-all system, which, however, fails to dynamically allocate inference resources based on the difficulty and domain of each query. To address this challenge, we shift away from the pursuit of a monolithic agentic system, instead optimizing the \textbf{agentic supernet}, a probabilistic and continuous distribution of agentic architectures. We introduce MaAS, an automated framework that samples query-dependent agentic systems from the supernet, delivering high-quality solutions and tailored resource allocation (\textit{e.g.}, LLM calls, tool calls, token cost). Comprehensive evaluation across six benchmarks demonstrates that MaAS \textbf{(I)} requires only $6\sim45\%$ of the inference costs of existing handcrafted or automated multi-agent systems, \textbf{(II)} surpasses them by $0.54\%\sim11.82\%$, and \textbf{(III)} enjoys superior cross-dataset and cross-LLM-backbone transferability.
format	Preprint
id	arxiv_https___arxiv_org_abs_2502_04180
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	Multi-agent Architecture Search via Agentic Supernet Zhang, Guibin Niu, Luyang Fang, Junfeng Wang, Kun Bai, Lei Wang, Xiang Machine Learning Computation and Language Multiagent Systems Large Language Model (LLM)-empowered multi-agent systems extend the cognitive boundaries of individual agents through disciplined collaboration and interaction, while constructing these systems often requires labor-intensive manual designs. Despite the availability of methods to automate the design of agentic workflows, they typically seek to identify a static, complex, one-size-fits-all system, which, however, fails to dynamically allocate inference resources based on the difficulty and domain of each query. To address this challenge, we shift away from the pursuit of a monolithic agentic system, instead optimizing the \textbf{agentic supernet}, a probabilistic and continuous distribution of agentic architectures. We introduce MaAS, an automated framework that samples query-dependent agentic systems from the supernet, delivering high-quality solutions and tailored resource allocation (\textit{e.g.}, LLM calls, tool calls, token cost). Comprehensive evaluation across six benchmarks demonstrates that MaAS \textbf{(I)} requires only $6\sim45\%$ of the inference costs of existing handcrafted or automated multi-agent systems, \textbf{(II)} surpasses them by $0.54\%\sim11.82\%$, and \textbf{(III)} enjoys superior cross-dataset and cross-LLM-backbone transferability.
title	Multi-agent Architecture Search via Agentic Supernet
topic	Machine Learning Computation and Language Multiagent Systems
url	https://arxiv.org/abs/2502.04180