Reproduce “Can Reinforcement Learning Find Stackelberg-Nash Equilibria in General-Sum Markov Games with Myopic Followers?” 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
Han Zhong, Zhuoran Yang, Zhaoran Wang, Michael I. Jordan
The authors develop two reinforcement-learning methods to find Stackelberg–Nash equilibria in sequential games with myopic followers and unknown linear transitions. They estimate the leader’s long-term value while modeling followers’ immediate Nash responses, using optimism online and pessimism offline. Online cumulative regret is bounded by tilde O(H²d³ᐟ²√K); offline rates require coverage of the target equilibrium trajectory, and multiple-follower games require a computational oracle.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.