Xin Liu

Assistant Professor, ShanghaiTech University

1 paper at NeurIPS 2025

OpenReview· Semantic Scholar· Google Scholar

Poster Session 2

1 paper

Wednesday, December 3, 2025 · 4:30 PM → 7:30 PM

Exhibit Hall C,D,E

Enhancing Safety in Reinforcement Learning with Human Feedback via Rectified Policy Optimization

#416 · Xiyue Peng, Hengquan Guo, jiawei zhang, Dongqing Zou, Ziyu Shao, Honghao Wei, Xin Liu

We propose Rectified Policy Optimization (RePO) to mitigate "safety compensation", which replaces the average safety metric with stricter safety constraints.