Fei-Yue Wang

Stop Summation: Min-Form Credit Assignment Is All Process Reward Model Needs for Reasoning

RIME: Robust Preference-based Reinforcement Learning with Noisy Human Preferences