Abstract: Reinforcement learning (RL) requires a lot of interactions with the environment, which is usually expensive or dangerous in real-world tasks. To address this problem, offline RL considers ...