Title:Two Timescale EXTRA for Smooth Non-convex Distributed Optimization Problems

Abstract:Distributed non-convex optimization over multi-agent networks is a growing area of interest. In this paper, we propose a decentralized algorithm called Two Time Scale EXTRA (TT-EXTRA), which can be considered as a modified version of the well-known EXTRA algorithm. EXTRA is very general and is closely related to gradient tracking-based algorithms, such as DIGGING, as well as the Primal-Dual Gradient algorithm in distributed settings. It has been established that EXTRA achieves sublinear convergence to an exact minimizer in the convex case and linear convergence in the strongly convex case. However, a convergence analysis of EXTRA for non-convex scenarios remains absent in the current literature. We address this gap by proving that TT-EXTRA converges to a set of consensual first-order stationary points for non-convex distributed optimization problems.