Reproduce “Provably Efficient Representation Learning for Low-Rank CMDPs” 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
Kaixuan Liu, Guojun Xiong, Shengpu Tang, Wanyun Si, Jian Li
MFRLC jointly learns state–action features and a reward–safety policy from hidden observations in constrained low-rank decision processes. It combines representation learning, optimistic value estimation, and primal–dual updates. Its theory bounds regret and constraint violation under realizability and Slater’s condition. On the H=6 CLCL benchmark, reward approaches the optimum and cost ends below threshold; the tested representation learner is outside the theorem.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.