Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Finite Sample Analysis of Minimax Offline Reinforcement Learning: Completeness, Fast Rates and First-Order Efficiency
(Submitted on 5 Feb 2021 (v1), last revised 24 Jul 2022 (this version, v2))
Abstract: We offer a theoretical characterization of off-policy evaluation (OPE) in reinforcement learning using function approximation for marginal importance weights and $q$-functions when these are estimated using recent minimax methods. Under various combinations of realizability and completeness assumptions, we show that the minimax approach enables us to achieve a fast rate of convergence for weights and quality functions, characterized by the critical inequality \citep{bartlett2005}. Based on this result, we analyze convergence rates for OPE. In particular, we introduce novel alternative completeness conditions under which OPE is feasible and we present the first finite-sample result with first-order efficiency in non-tabular environments, i.e., having the minimal coefficient in the leading term.
Submission history
From: Masatoshi Uehara [view email][v1] Fri, 5 Feb 2021 03:20:39 GMT (137kb)
[v2] Sun, 24 Jul 2022 23:49:47 GMT (228kb)
Link back to: arXiv, form interface, contact.