24th Conference of the International Federation of Operational Research Societies
Abstract Submission

1699. A Unified Decentralized Nonconvex Algorithm under Kurdyka-Lojasiewicz Condition

Invited abstract in session MC-37: Optimization algorithms and applications 1, stream Nonlinear Optimization.

Monday, 10:30-12:00
Room: JUR – Seminar-Raum 43

Authors (first author is the speaker)

1. Liping Wang
School of Mathematics, Nanjing University of Aeronautics and Astronautics

Abstract

In this paper, a unified decentralized nonconvex algorithmic framework is proposed which subsumes several state-of-the-art gradient tracking algorithms and quasi-Newton algorithms. We develop an analytical convergence of the unified algorithm under the Kurdyka-Lojasiewicz condition for nonconvex optimization problem. We also propose some quasi-Newton variants that fit into the proposed framework. The numerical results show that these newly developed algorithms are very efficient compared with other state-of-the-art algorithms for solving decentralized nonconvex smooth optimization problems.

Keywords

Status: accepted


Back to the list of papers