Reproduce “Entropy-regularized Point-based Value Iteration” 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
Harrison Delecki, Marcell Vazquez-Chanlatte, Esen Yel, Kyle Wray, Tomer Arnon, Stefan Witwicki, Mykel J. Kochenderfer
The authors propose ERPBVI, an offline approximate solver for partially observable Markov decision processes that adds an entropy bonus to action values and makes policies less committed. They compare it with PBVI in robustness and goal-inference tests. ERPBVI improves mean return by up to 22.62 in one tested Tiger observation mismatch, but not when test observations are more accurate than training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.