Qisai Liu
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Proximal Preference Optimization: A Unified Framework for Stable and Reward-Free Alignment
According to the abstract, Qisai Liu and colleagues combine the loss of Direct Preference Optimization with the trust-region idea of PPO in two new algorithms, P2O and PKTO, and report they match or exceed DPO and KTO in alignment quality with better stability.