Secure cloud computing algorithms for discrete constrained potential games∗

In this paper, we study secure cloud computing problem for a class of discrete constrained potential games. In the games, certain functions are confidential for the system operator and not disclosed to any other participant. Meanwhile, each agent is unwilling to disclose its private functions and states to any other participant. By utilizing reinforcement learning and homomorphic encryption, we propose a distributed algorithm where (i) both the confidentiality for the system operator and the privacy for the agents are protected; (ii) the convergence to Nash equilibria is formally ensured.