Reproduce “Dynamic Policy Programming” 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

Mohammad Gheshlaghi Azar, Vicenc Gomez, Hilbert J. Kappen

DPP learns policies by carrying accumulated update information forward rather than relying only on the latest estimate. Its approximate performance-loss bound depends on average accumulated approximation error. On the combination lock, DPP-RL’s mean error was 0.20 versus 18.18 for tested Q-learning after 30 seconds, using a generative model; this result does not establish superiority beyond the tested simulated benchmarks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.