Data-Incremental Continual Offline Reinforcement Learning

Sibo Gai, Donglin Wang

In this work, we propose a new setting of continual learning: data-incremental continual offline reinforcement learning (DICORL), in which an agent is asked to learn a sequence of…