Privacy-preserving Decision Making Based on Q-Learning in Cloud Computing

People encounter a variety of continuous decision-making (DM) problems in the real world. Reinforcement learning (RL) is a promising technique to solve these problems. This paper proposes a privacy-preserving Q-learning decision-making scheme (PQDM). Based on distributed homomorphic encryption (HE), we design several secure protocols to implement the underlying nonlinear operations such as comparing, maximizing, and maximizing parameter solving. Based on the designed security protocols, we propose a secure decision-making protocol in cloud computing, which enables the cloud server to perform element selection and Q-learning functions on ciphertext data. During the entire process, the cloud server does not need to know the actual state, thus guaranteeing the security of the original state information. We analyze the security and complexity of the whole scheme theoretically. Our experimental results show our proposed scheme's effectiveness and good spatio-temporal performance.