InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modeling

TL;DR

This paper introduces InfoRM, an information-theoretic framework for reward modeling in RLHF that reduces reward hacking by filtering irrelevant information and detecting overoptimization through latent space analysis.

Contribution

It proposes a variational information bottleneck approach for reward modeling and introduces the Cluster Separation Index for online detection of reward overoptimization.

Findings

InfoRM effectively mitigates reward hacking across various scales.

The IB latent space correlates with reward overoptimization, enabling detection.

The Cluster Separation Index reliably indicates reward overoptimization in diverse datasets.

Abstract

Despite the success of reinforcement learning from human feedback (RLHF) in aligning language models with human values, reward hacking, also termed reward overoptimization, remains a critical challenge. This issue primarily arises from reward misgeneralization, where reward models (RMs) compute reward using spurious features that are irrelevant to human preferences. In this work, we tackle this problem from an information-theoretic perspective and propose a framework for reward modeling, namely InfoRM, by introducing a variational information bottleneck objective to filter out irrelevant information. Notably, we further identify a correlation between overoptimization and outliers in the IB latent space of InfoRM, establishing it as a promising tool for detecting reward overoptimization. Inspired by this finding, we propose the Cluster Separation Index (CSI), which quantifies deviations in the IB latent space, as an indicator of reward overoptimization to facilitate the development of online mitigation strategies. Extensive experiments on a wide range of settings and RM scales (70M, 440M, 1.4B, and 7B) demonstrate the effectiveness of InfoRM. Further analyses reveal that InfoRM's overoptimization detection mechanism is not only effective but also robust across a broad range of datasets, signifying a notable advancement in the field of RLHF. The code will be released upon acceptance.