1 paper · 1 filter
Jiaxin Zhao, Weihang Pan, Xun Liang +1
Offline policy improvement faces an inherent conflict between maximizing value and fitting the data distribution. While in-sample weighted regression is stable, it suffers from ove…