Reproduce “Optimal Transport for LLM Reward Modeling from Noisy Preference” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Licheng Pan, Haochen Yang, Haoxuan Li, Yunsheng Lu, Yongqi Tong, Yinuo Wang, Shijian Wang, Zhixuan Chu and 3 more
SelectiveRM trains reward models by combining semantic distance with preference loss in a partial optimal-transport plan, leaving high-cost examples unmatched rather than fitting every label. On PKU-SafeRLHF with 20% simulated label flips, it reduces test MSE from 0.072 to 0.057 when partial transport is added to the joint-cost method. This supports selection in that setup, not naturally noisy human preferences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.