A Compressed Gradient Tracking Method for Decentralized Optimization With Linear Convergence

Communication compression techniques are of growing interests for solving the decentralized optimization problem under limited communication, where global objective is to minimize average local cost functions over a multiagent network using only computation and peer-to-peer communication. In this article, we propose novel compressed gradient tracking algorithm (C-GT) that combines technique with communication compression. particular, C-GT compatible general class operators unifies both unbiased biased compressors. We show inherits advantages tracking-based algorithms achieves linear convergence rate strongly convex smooth functions. Numerical examples complement theoretical findings demonstrate efficiency flexibility proposed algorithm.