Mitigating Reward Over-Optimization in RLHF via Behavior-Supported Regularization.

Juntao Dai, Taiye Chen, Yaodong Yang, Qian Zheng, Gang Pan

Browse the full ICLR paper archive.