Reproduce “Towards Blackwell Optimality: Bellman Optimality Is All You Can Get” 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
Victor Boone, Adrienne Tuynman
HOPE learns policies in communicating Markov decision processes across finite optimality orders by repeatedly constructing empirical models and using epsilon-tolerant higher-order policy iteration. Its recommendation error vanishes for each finite order. However, finite-time stopping with a confidence guarantee is possible exactly when the MDP has a unique Bellman-optimal policy at every target order; this does not rule out asymptotic learning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.