Reproduce “Offline Bayesian Aleatoric and Epistemic Uncertainty Quantification and Posterior Value Optimisation in Finite-State 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
Filippo Valdettaro, A. Aldo Faisal
The authors develop an offline reinforcement learning method that separates uncertainty about transition models from randomness within each model. It computes return moments analytically for posterior-sampled finite-state Markov decision processes and optimizes policy value across plausible models. On MIMIC-III data, it raises model-estimated survival value about 2.1% on average under a conservative prior, not measured survival.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.