Reproduce “FLAMBE: Structural Complexity and Representation Learning of Low Rank MDPs” 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
Alekh Agarwal, Sham Kakade, Akshay Krishnamurthy, Wen Sun
FLAMBE addresses representation learning and exploration in rich-state reinforcement learning by iteratively collecting transitions, fitting a low-rank transition model, and using it to guide further exploration. Under realizability assumptions, it learns a model with bounded expected one-step total-variation error under each policy’s induced distribution. The trajectory bound is theoretical and relies on maximum-likelihood and sampling oracles.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.