Saved in:
Bibliographic Details
Main Authors: Zhao, Wanjia, Yuksekgonul, Mert, Wu, Shirley, Zou, James
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.04780
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929702266994688
author Zhao, Wanjia
Yuksekgonul, Mert
Wu, Shirley
Zou, James
author_facet Zhao, Wanjia
Yuksekgonul, Mert
Wu, Shirley
Zou, James
contents Multi-agent AI systems powered by large language models (LLMs) are increasingly applied to solve complex tasks. However, these systems often rely on fragile, manually designed prompts and heuristics, making optimization difficult. A key challenge in optimizing multi-agent systems is acquiring suitable training data for specialized agents. We introduce SiriuS, a self-improving, reasoning-driven optimization framework for multi-agent systems. Central to our approach is the construction of an experience library: a repository of high-quality reasoning trajectories. The library is built by retaining reasoning steps that lead to successful outcomes, providing a robust training set for optimizing multi-agent system. Additionally, we introduce a library augmentation procedure that refines unsuccessful trajectories, further enriching the library. SiriuS boosts performance by 2.86\% to 21.88\% on reasoning and biomedical QA and enhances agent negotiation in competitive settings. Our results show that SiriuS enhances multi-agent performance while generating reusable data for self-correction and self-play enhancement in the future.
format Preprint
id arxiv_https___arxiv_org_abs_2502_04780
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SiriuS: Self-improving Multi-agent Systems via Bootstrapped Reasoning
Zhao, Wanjia
Yuksekgonul, Mert
Wu, Shirley
Zou, James
Artificial Intelligence
Multi-agent AI systems powered by large language models (LLMs) are increasingly applied to solve complex tasks. However, these systems often rely on fragile, manually designed prompts and heuristics, making optimization difficult. A key challenge in optimizing multi-agent systems is acquiring suitable training data for specialized agents. We introduce SiriuS, a self-improving, reasoning-driven optimization framework for multi-agent systems. Central to our approach is the construction of an experience library: a repository of high-quality reasoning trajectories. The library is built by retaining reasoning steps that lead to successful outcomes, providing a robust training set for optimizing multi-agent system. Additionally, we introduce a library augmentation procedure that refines unsuccessful trajectories, further enriching the library. SiriuS boosts performance by 2.86\% to 21.88\% on reasoning and biomedical QA and enhances agent negotiation in competitive settings. Our results show that SiriuS enhances multi-agent performance while generating reusable data for self-correction and self-play enhancement in the future.
title SiriuS: Self-improving Multi-agent Systems via Bootstrapped Reasoning
topic Artificial Intelligence
url https://arxiv.org/abs/2502.04780