Distributed Online Bandit Optimization with Communication Compression

This paper is concerned with a distributed online optimization issue in a network consisting of multiple interacting nodes. In order to alleviate the communication burden, we propose a communication-efficient distributed online optimization algorithm that utilizes bandit feedback and communication compression over an undirected graph. The convergence of the algorithm is analyzed theoretically, from which we give the regret bound, and some numerical results are provided to demonstrate the effectiveness of the algorithm.