English

Improving Reinforcement Learning from Human Feedback with Efficient Reward Model Ensemble

Machine Learning 2024-10-23 v3 Artificial Intelligence Computation and Language

Abstract

Reinforcement Learning from Human Feedback (RLHF) is a widely adopted approach for aligning large language models with human values. However, RLHF relies on a reward model that is trained with a limited amount of human preference data, which could lead to inaccurate predictions. As a result, RLHF may produce outputs that are misaligned with human values. To mitigate this issue, we contribute a reward ensemble method that allows the reward model to make more accurate predictions. As using an ensemble of large language model-based reward models can be computationally and resource-expensive, we explore efficient ensemble methods including linear-layer ensemble and LoRA-based ensemble. Empirically, we run Best-of-nn and Proximal Policy Optimization with our ensembled reward models, and verify that our ensemble methods help improve the alignment performance of RLHF outputs.

Keywords

Cite

@article{arxiv.2401.16635,
  title  = {Improving Reinforcement Learning from Human Feedback with Efficient Reward Model Ensemble},
  author = {Shun Zhang and Zhenfang Chen and Sunli Chen and Yikang Shen and Zhiqing Sun and Chuang Gan},
  journal= {arXiv preprint arXiv:2401.16635},
  year   = {2024}
}
R2 v1 2026-06-28T14:31:00.374Z