Policy Finetuning in Reinforcement Learning via Design of Experiments using Offline Data

TL;DR

This paper introduces a theoretically grounded method for designing a single exploration policy in reinforcement learning that effectively leverages offline data and minimal online exploration to improve policy quality.

Contribution

It proposes an algorithm with provable guarantees for offline data-driven exploration policy design, analyzing its performance based on dataset coverage and additional data.

Findings

Algorithm has provable guarantees.

Performance depends on dataset coverage.

Effective with limited online data.

Abstract

In some applications of reinforcement learning, a dataset of pre-collected experience is already available but it is also possible to acquire some additional online data to help improve the quality of the policy. However, it may be preferable to gather additional data with a single, non-reactive exploration policy and avoid the engineering costs associated with switching policies. In this paper we propose an algorithm with provable guarantees that can leverage an offline dataset to design a single non-reactive policy for exploration. We theoretically analyze the algorithm and measure the quality of the final policy as a function of the local coverage of the original dataset and the amount of additional data collected.