Information-Theoretic Reward Modeling for Stable RLHF: Detecting and Mitigating Reward Hacking
Comments 46 pages, 36 figures, submitted to IEEE Transactions on Pattern Analysis and Machine Intelligence
期刊&会议
IEEE Transactions on Pattern Analysis and Machine Intelligence · 期刊 · Computer Vision
Comments 46 pages, 36 figures, submitted to IEEE Transactions on Pattern Analysis and Machine Intelligence