Policy Gradient Primal-Dual Method for Safe Reinforcement Learning from Human Feedback

TL;DR

This paper introduces a novel policy gradient primal-dual approach for safe reinforcement learning from human feedback, modeling it as an infinite horizon discounted constrained MDP with convergence guarantees.

Contribution

It formulates safe RLHF as an infinite horizon discounted CMDP and proposes two algorithms that do not require reward model fitting, supporting flexible trajectories with proven convergence.

Findings

First to study infinite horizon discounted CMDP under human feedback

Algorithms achieve global convergence with polynomial rates

Supports flexible trajectory lengths for training

Abstract

Safe Reinforcement Learning from Human Feedback (Safe RLHF) has recently achieved empirical success in developing helpful and harmless large language models by decoupling human preferences regarding helpfulness and harmlessness. Existing approaches typically rely on fitting fixed horizon reward models from human feedback and have only been validated empirically. In this paper, we formulate safe RLHF as an infinite horizon discounted Con- strained Markov Decision Process (CMDP), since humans may interact with the model over a continuing sequence of interactions rather than within a single finite episode. We propose two Safe RLHF algorithms that do not require reward model fitting and, in contrast to prior work assuming fixed-length trajectories, support flexible trajectory lengths for training. Both algo- rithms are based on the primal-dual method and achieve global convergence guarantees with polynomial rates in terms of policy gradient iterations, trajectory sample lengths, and human preference queries. To the best of our knowledge, this is the first work to study infinite horizon discounted CMDP under human feedback and establish global, non-asymptotic convergence.