Reproduce “Minimax Optimal Strategy for Delayed Observations in Online Reinforcement Learning” 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
Harin Lee, Kevin Jamieson
The authors model delayed state observation as an augmented Markov decision process and develop an optimistic method that shares transition estimates across states. In tabular settings, it achieves an upper regret bound of \(\tilde O(H\sqrt{D_{\max}SAK})\) and matches a lower bound up to logarithmic factors under stated conditions, while runtime remains exponential in maximum delay.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.