Offline Quantum Reinforcement Learning in a Conservative Manner

Recently, to reap the quantum advantage, empowering reinforcement learning (RL) with computing has attracted much attention, which is dubbed as RL (QRL). However, current QRL algorithms employ an online scheme, i.e., policy that run on a computer needs interact environment collect experiences, could be expensive and dangerous for practical applications. In this paper, we aim solve problem in offline manner. To more specific, develop first (offline QRL) algorithm named CQ2L (Conservative Quantum Q-learning), learns from samples does not require any interaction environment. utilizes variational circuits (VQCs), are improved data re-uploading scaling parameters, represent Q-value functions of agents. suppress overestimation Q-values resulting data, double Q-learning framework reduce bias; then penalty term encourages generating conservative designed. We conduct abundant experiments demonstrate proposed method can successfully tasks counterpart not.