Reproduce “TacoMAS: Test-Time Co-Evolution of Topology and Capability in LLM-based Multi-Agent Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chen Xu, Yicheng Hu, Ruizi Wang, Xinyu Lin, Wenjie Wang, Dongrui Liu, Fuli Feng
TacoMAS introduces a framework for LLM-based multi-agent systems that co-evolves agent capabilities and communication topology during test-time inference using distinct timescales. This approach yields an average accuracy improvement of 13.3% across four diverse benchmarks compared to leading baselines.
Key insights
- Optimal MAS performance requires joint adaptation of both agent capabilities and communication topology.
- Separating the timescales of adaptation (fast for capabilities, slow for topology) is crucial for maintaining system stability and achieving effective co-evolution.
- The meta-LLM can serve as an effective diagnostic coach to drive both localized agent refinements and global structural reorganizations in a principled manner.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.