Skip to yearly menu bar Skip to main content


Poster
in
Workshop: Mathematical and Empirical Understanding of Foundation Models (ME-FoMo)

Principled Reinforcement Learning with Human Feedback from Pairwise or $K$-wise Comparisons

Banghua Zhu · Jiantao Jiao · Michael Jordan

Keywords: [ Pessimism ] [ reinforcement learning with human feedback (RLHF) ] [ Plackett-Luce model ] [ Bradley-Terry- Luce model ] [ maximum likelihood estimator ] [ offline reinforcement learning ]


Abstract: We provide a theoretical framework for Reinforcement Learning with Human Feedback (RLHF). Our analysis shows that when the true reward function is linear, the widely used maximum likelihood estimator (MLE) converges under both the Bradley-Terry-Luce (BTL) model and the Plackett-Luce (PL) model. However, we show that when training a policy based on the learned reward model, MLE fails while a pessimistic MLE provides policies with improved performance under certain coverage assumptions. Additionally, we demonstrate that under the PL model, the true MLE and an alternative MLE that splits the $K$-wise comparison into pairwise comparisons both converge. Moreover, the true MLE is asymptotically more efficient. Our results validate the empirical success of existing RLHF algorithms in InstructGPT and provide new insights for algorithm design. Furthermore, our results unify the problem of RLHF and max entropy Inverse Reinforcement Learning (IRL), and provide the first sample complexity bound for max entropy IRL.

Chat is not available.