Reproduce “Conflux-PSRO: Effectively Leveraging Collective Advantages in Policy Space Response Oracles” 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
Yucong Huang, Jiesong Lian, Mingzhi Wang, Chengdong Ma, Ying Wen
Conflux-PSRO, from researchers at Peking University and Shanghai Jiao Tong University, improves Best Response generation within the Policy Space Response Oracle (PSRO) framework for zero-sum games. It achieves this by intelligently combining diverse existing policies via a Mixture of Experts architecture and distillation, leading to reduced exploitability across various poker and board games.
Key insights
- Instead of blindly seeking diversity, leveraging the collective advantages of useful diverse policies can create stronger Best Responses.
- Dynamically combining specialized policies at a fine-grained, state-level can outperform static or episode-level policy mixtures.
- Distilling complex ensemble behaviors into a single policy maintains compatibility with iterative frameworks like PSRO while retaining performance benefits.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.