Exploration-Driven Policy Optimization in RLHF: Theoretical Insights on Efficient Data Utilizationby Yihan Du, Anna Winnicki,…
Reward Generalization in RLHF: A Topological Perspectiveby Tianyi Qiu, Fanzhi Zeng, Jiaming Ji, Dong Yan,…
RS-DPO: A Hybrid Rejection Sampling and Direct Preference Optimization Method for Alignment of Large Language…
InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modelingby Yuchun Miao, Sen Zhang, Liang…
Reinforcement Learning from Human Feedback with Active Queriesby Kaixuan Ji, Jiafan He, Quanquan GuFirst submitted…
MaxMin-RLHF: Alignment with Diverse Human Preferencesby Souradip Chakraborty, Jiahao Qiu, Hui Yuan, Alec Koppel, Furong…
Active Preference Learning for Large Language Modelsby William Muldrew, Peter Hayes, Mingtian Zhang, David BarberFirst…
Online Iterative Reinforcement Learning from Human Feedback with General Preference Modelby Chenlu Ye, Wei Xiong,…
ODIN: Disentangled Reward Mitigates Hacking in RLHFby Lichang Chen, Chen Zhu, Davit Soselia, Jiuhai Chen,…
Corruption Robust Offline Reinforcement Learning with Human Feedbackby Debmalya Mandal, Andi Nika, Parameswaran Kamalaruban, Adish…