TL;DR
EvoMAS is an evolutionary approach to automatically generate multi-agent systems that outperform prior methods in robustness, executability, and task performance across diverse benchmarks.
Contribution
EvoMAS introduces a structured, evolutionary configuration generation method for MAS, improving robustness and performance over existing automatic and human-designed systems.
Findings
EvoMAS outperforms prior methods on reasoning and tool-use benchmarks.
EvoMAS achieves 79.1% on SWE-Bench-Verified, matching top leaderboard scores.
EvoMAS improves task performance by +10.5 and +7.1 points over EvoAgent.
Abstract
Large language model (LLM)-based multi-agent systems (MAS) show strong promise for complex reasoning, planning, and tool-augmented tasks, but designing effective MAS architectures remains labor-intensive, brittle, and hard to generalize. Existing automatic MAS generation methods either rely on code generation, which often leads to executability and robustness failures, or impose rigid architectural templates that limit expressiveness and adaptability. We propose Evolutionary Generation of Multi-Agent Systems (EvoMAS), which formulates MAS generation as structured configuration generation. EvoMAS performs evolutionary generation in configuration space. Specifically, EvoMAS selects initial configurations from a pool, applies feedback-conditioned mutation and crossover guided by execution traces, and iteratively refines both the candidate pool and an experience memory. We evaluate EvoMAS…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsMulti-Agent Systems and Negotiation · Reinforcement Learning in Robotics · Language and cultural evolution
