Offline Reinforcement Learning via Linear-Programming with Error-Bound Induced Constraints

TL;DR

This paper develops a linear programming approach for offline reinforcement learning that achieves optimal sample complexity by incorporating error bounds as constraints, applicable to various MDP settings and function approximation methods.

Contribution

It introduces a novel LP framework with error-bound constraints that attain optimal sample complexity under general conditions, relaxing previous assumptions.

Findings

Achieves $O(1/\sqrt{n})$ sample complexity with partial data coverage.

Handles both infinite-horizon discounted and average-reward MDPs.

Provides state-of-the-art or first sample complexities in tabular offline RL settings.

Abstract

Offline reinforcement learning (RL) aims to find an optimal policy for Markov decision processes (MDPs) using a pre-collected dataset. In this work, we revisit the linear programming (LP) reformulation of Markov decision processes for offline RL, with the goal of developing algorithms with optimal $O(1/\sqrt{n})$ sample complexity, where $n$ is the sample size, under partial data coverage and general function approximation, and with favorable computational tractability. To this end, we derive new \emph{error bounds} for both the dual and primal-dual formulations of the LP, and incorporate them properly as \emph{constraints} in the LP reformulation. We then show that under a completeness-type assumption, $O(1/\sqrt{n})$ sample complexity can be achieved under standard single-policy coverage assumption, when one properly \emph{relaxes} the occupancy validity constraint in the LP. This framework can readily handle both infinite-horizon discounted and average-reward MDPs, in both general function approximation and tabular cases. The instantiation to the tabular case achieves either state-of-the-art or the first sample complexities of offline RL in these settings. To further remove any completeness-type assumption, we then introduce a proper \emph{lower-bound constraint} in the LP, and a variant of the standard single-policy coverage assumption. Such an algorithm leads to a $O(1/\sqrt{n})$ sample complexity with dependence on the \emph{value-function gap}, with only realizability assumptions. Our properly constrained LP framework advances the existing results in several aspects, in relaxing certain assumptions and achieving the optimal $O(1/\sqrt{n})$ sample complexity, with simple analyses. We hope our results bring new insights into the use of LP formulations and the equivalent primal-dual minimax optimization for offline RL, through the error-bound induced constraints.