The Least Restriction for Offline Reinforcement Learning

Zizhou Su

Many practical applications of reinforcement learning (RL) constrain the agent to learn from a fixed offline dataset of logged interactions, which has already been gathered, withou…