Skip to main navigation Skip to search Skip to main content

Distributed Mirror Descent for Nonconvex Constrained Optimization

  • Wei Suo
  • , Wenling Li*
  • *Corresponding author for this work
  • Beihang University

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This paper is concerned with a class of distributed online constrained optimization problems characterized by several key features: i) the complex coupling characteristics of multiple coupled constraints; ii) the dynamic unbalance of time-varying (TV) digraphs; iii) the nonconvex nature of local cost functions. To tackle these intricate challenges effectively, a primal dual proximal mirror descent (PDPMD) algorithm is developed. Furthermore, an auxiliary variable is employed to counteract the imbalance induced by TV directed graphs. Additionally, we prove that the proposed method, under some mild conditions, reaches stationary points with a sublinear convergence rate. At last, a numerical example is used to illustrate the validity of the proposed algorithm.

Original languageEnglish
Title of host publicationIntelligent Networked Things - 8th China Intelligent Networked Things Conference, CINT 2025, Proceedings
EditorsLin Zhang, Yuanjun Laili, Wensheng Yu, Ting Qu
PublisherSpringer Science and Business Media Deutschland GmbH
Pages298-306
Number of pages9
ISBN (Print)9789819511020
DOIs
StatePublished - 2026
Event8th China Intelligent Networked Things Conference, CINT 2025 - Zhuhai, China
Duration: 13 Jun 202515 Jun 2025

Publication series

NameCommunications in Computer and Information Science
Volume2624 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference8th China Intelligent Networked Things Conference, CINT 2025
Country/TerritoryChina
CityZhuhai
Period13/06/2515/06/25

Keywords

  • Distributed nonconvex optimization
  • Multiple coupled constraints
  • Online learning
  • Time-varying unbalanced digraphs

Fingerprint

Dive into the research topics of 'Distributed Mirror Descent for Nonconvex Constrained Optimization'. Together they form a unique fingerprint.

Cite this