Reproduce “DiffNash: Finding Nash Equilibrium in High-Dimensional Continuous Action Spaces with Diffusion Policies” 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
Haichuan Wang, Anagha Satish, Jiale Zheng, Saisneha Ghatti, Lingkai Kong, Kai Wang, Milind Tambe
DiffNash represents players’ mixed strategies in continuous games with diffusion models and updates them using Online Mirror Descent. It trains with utility-weighted losses and reuses past actions through replay correction based on an Evidence Lower Bound proxy. Across six forest configurations, it reduces estimated Nash gap by 35.6% on average versus the strongest baseline; this uses approximate best responses.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.