Reinforcement learning approach to shortcuts between thermodynamic states with minimum entropy production

We propose a systematic method based on reinforcement learning (RL) techniques to find the optimal path to minimize the total entropy production between two equilibrium states of open systems at the same temperature in a given fixed period. Benefiting from the generalization of the deep RL techniques, we provide a powerful tool to address this problem in quantum systems even with two-dimensional continuous controllable parameters. We successfully apply our method to the classical and quantum two-level systems.