Binary Reward Labeling: Bridging Offline Preference and Reward-Based Reinforcement Learning

TL;DR

This paper introduces a framework that converts preference feedback into scalar rewards using binary reward labeling, enabling the application of reward-based offline RL algorithms to preference-based data, thus bridging a significant gap in offline RL research.

Contribution

The authors propose a universal framework that transforms preference feedback into scalar rewards, allowing existing reward-based offline RL algorithms to be used for preference-based data, with theoretical and empirical validation.

Findings

Framework achieves comparable results to reward-based RL on benchmark datasets.

Binary reward labeling minimizes information loss during feedback transition.

Combining reward labeling with various algorithms improves offline preference RL performance.

Abstract

Offline reinforcement learning has become one of the most practical RL settings. However, most existing works on offline RL focus on the standard setting with scalar reward feedback. It remains unknown how to universally transfer the existing rich understanding of offline RL from the reward-based to the preference-based setting. In this work, we propose a general framework to bridge this gap. Our key insight is transforming preference feedback to scalar rewards via binary reward labeling (BRL), and then any reward-based offline RL algorithms can be applied to the dataset with the reward labels. The information loss during the feedback signal transition is minimized with binary reward labeling in the practical learning scenarios. We theoretically show the connection between several recent PBRL techniques and our framework combined with specific offline RL algorithms. By combining reward labeling with different algorithms, our framework can lead to new and potentially more efficient offline PBRL algorithms. We empirically test our framework on preference datasets based on the standard D4RL benchmark. When combined with a variety of efficient reward-based offline RL algorithms, the learning result achieved under our framework is comparable to training the same algorithm on the dataset with actual rewards in many cases and better than the recent PBRL baselines in most cases.