Representation Matters: Offline Pretraining for Sequential Decision Making

                         Mengjiao Yang 1 Ofir Nachum 1

              Abstract
    The recent success of supervised learning meth-
    ods on ever larger offline datasets has spurred
    interest in the reinforcement learning (RL) field
    to investigate whether the same paradigms can
    be translated to RL algorithms. This research
    area, known as offline RL, has largely focused
    on offline policy optimization,  ...