Distributed Stochastic Optimization with Gradient Tracking over Time-Varying Directed Networks

arXiv2305.00629MaRDI QIDQ6434858FDOQ6434858

Authors: Duong Thuy Anh Nguyen, Duong Tung Nguyen, Angelia Nedić 

Publication date: 30 April 2023

Abstract: We propose a novel distributed method called SAB-TV that employs gradient tracking for minimizing the sum of smooth and strongly-convex local cost functions of networked agents communicating over a time-varying directed graph. Assuming access to a stochastic first-order oracle (SFO), each agent uses an auxiliary variable to asymptotically track the gradient of the global cost in expectation. The SAB-TV algorithm utilizes row- and column-stochastic weights to ensure both consensus and optimality, and can be applied to arbitrary, strongly-connected directed, and time-varying communication graphs. We show that, with a sufficiently small constant step-size, SAB-TV converges linearly (in expected mean-square sense) to a neighborhood of the global minimizer. To illustrate the theoretical results, we have conducted numerical simulations based on real-world datasets.

This page was built for publication: Distributed Stochastic Optimization with Gradient Tracking over Time-Varying Directed Networks

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6434858)