Reproduce “Reinforcement Learning Under Runtime Confounding: Policy Learning and Off-Policy Evaluation” 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
Raphael Kim, Michele Santacatterina, Ramin Zabih, Keith Barnatchez
The authors develop a framework for evaluating and learning sequential policies when logged variables are unavailable at deployment. The target policy uses runtime-visible history, while action propensities are estimated from full historical records to account for past assignments. Under causal identification and estimation conditions, they derive an efficient off-policy estimator and a policy-learning regret bound; results are theoretical, not experimental.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.