Reproduce “From Single-Agent Self-Discovery to Multi-Agent Adaptation: Scaling Laws of Partner Feedback and Convention Rigidity in Meta Reinforcement Learning Algorithms” 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
Haomin Luo, Ersha Li
The authors compare learned reinforcement-learning update rules and feedback routes in cooperative multi-agent tasks. They assess trained pairs alongside focal agents evaluated with held-out partners, and vary physical coupling and training horizon. In Overcooked, coupled MAPPO earns higher self-play return, while coupled Disco103 has higher held-out-bank return and fewer collisions. Results depend on task, horizon, and synthetic partner bank.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.