1 paper
Miteto Wei, Xiaohan Wang, Zehao Chen +7
On-policy distillation (OPD) reduces train-test state mismatch by training a student on its own generated trajectories, but weak students may visit teacher-misaligned prefixes wher…