The Role of Inherent Bellman Error in Offline Reinforcement Learning with Linear Function Approximation

TL;DR

This paper investigates offline reinforcement learning with linear function approximation, demonstrating that low inherent Bellman error enables efficient algorithms under single-policy coverage, with suboptimality scaling as the square root of the Bellman error.

Contribution

It introduces a computationally efficient algorithm for offline RL under low inherent Bellman error and single-policy coverage, providing the first guarantees in the linear Bellman completeness setting.

Findings

Algorithm succeeds under single-policy coverage.

Suboptimality scales with the square root of Bellman error.

Lower bound shows this scaling cannot be improved.

Abstract

In this paper, we study the offline RL problem with linear function approximation. Our main structural assumption is that the MDP has low inherent Bellman error, which stipulates that linear value functions have linear Bellman backups with respect to the greedy policy. This assumption is natural in that it is essentially the minimal assumption required for value iteration to succeed. We give a computationally efficient algorithm which succeeds under a single-policy coverage condition on the dataset, namely which outputs a policy whose value is at least that of any policy which is well-covered by the dataset. Even in the setting when the inherent Bellman error is 0 (termed linear Bellman completeness), our algorithm yields the first known guarantee under single-policy coverage. In the setting of positive inherent Bellman error ${\varepsilon_{\mathrm{BE}}} > 0$, we show that the suboptimality error of our algorithm scales with $\sqrt{\varepsilon_{\mathrm{BE}}}$. Furthermore, we prove that the scaling of the suboptimality with $\sqrt{\varepsilon_{\mathrm{BE}}}$ cannot be improved for any algorithm. Our lower bound stands in contrast to many other settings in reinforcement learning with misspecification, where one can typically obtain performance that degrades linearly with the misspecification error.