A Distributed Algorithm with a Modified Quantization Radius Under Limited Communication

Abstract

Distributed optimization is a very significant approach with applications in control theory and lots of related fields, as it is high fault-tolerant and extremely efficient compared with centralized optimization. In distributed optimization, quantization is a communication technique, which exchanges information bits more reliably at the expense of a lower communication rate. In this paper, we put forward a distributed quantization algorithm to handle a class of problems with the sum of multiple objective functions. Particularly, we design a customized quantization radius, which preserves a fast convergence rate while saving more bits of information exchange. The numerical experiments demonstrate the high efficiency of our algorithm.