Reproduce “GoAgent: Group-of-Agents Communication Topology Generation for 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
Hongjiang Chen, Xin Zheng, Yixin Liu, Pengfei Jiao, Shiyuan Li, Huan Liu, Zhidong Zhao, Ziqi Xu and 2 more
GoAgent introduces a group-centric, autoregressive method for building multi-agent system communication topologies. It leverages a Conditional Information Bottleneck to reduce token consumption by 17% and improve accuracy by up to 2.47% on complex tasks while enhancing robustness against prompt injection attacks compared to previous state-of-the-art methods.
Key insights
- Explicitly modeling collaborative groups, rather than individual agents, as atomic units for multi-agent system construction significantly improves coordination and problem-solving for complex tasks.
- Incorporating a Conditional Information Bottleneck during topology generation is crucial for achieving token efficiency and enhancing robustness by filtering out task-irrelevant information and restricting error propagation.
- The optimal compression strategy for group prediction and edge prediction is asymmetric, with edge prediction benefiting more significantly from information compression to prevent redundant connections.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.