SiriuS: Self-improving Multi-agent Systems via Bootstrapped Reasoning

TL;DR

SiriuS is a self-improving multi-agent system framework that uses high-quality reasoning trajectories to optimize performance and facilitate self-correction, significantly improving reasoning and biomedical QA tasks.

Contribution

The paper introduces SiriuS, a novel framework that constructs and refines an experience library of reasoning trajectories to self-improve multi-agent systems without manual prompt engineering.

Findings

Boosts reasoning and biomedical QA performance by up to 21.88%

Enables self-correction and self-play through reusable data

Improves agent negotiation in competitive settings

Abstract

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.