Distributed Stochastic Optimization with Heavy-Ball Momentum Term for Parameter Estimation

Victoria Erofeeva, Oleg Granichin, Anna Sergeenko

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

Abstract

In this paper, we analyze DSPSA: a new distributed optimization algorithm for problems involving uncertainties. DSPSA combines Simultaneous perturbation stochastic approximation with consensus protocol and possesses properties of both algorithms. We study this method in the context of parameter estimation problems over large-scale sensor networks. Optimization in such networks may lead to communication overhead. This problem sets new requirements on optimization algorithms that must account for the efficacy of communication. Despite the presence of uncertainties: noise, external disturbance, and time-varying topology due to communication constraints, DSPSA converges to parameters to be estimated. The theoretical results provide an asymptotically efficient upper bound for the residuals. We also ananyze the convergence of the algorithm with the involvmenet of the heavy-ball momemnum term.

Original languageEnglish
Title of host publicationConference Proceedings - 5th Scientific School Dynamics of Complex Networks and their Applications, DCNA 2021
EditorsAlexander Hramov, Semen Kurkin, Andrey Andreev, Natalia Shusharina
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages69-72
Number of pages4
ISBN (Electronic)9781665442824
ISBN (Print)978-1-6654-4284-8
DOIs
StatePublished - 13 Sep 2021
Event5th Scientific School on Dynamics of Complex Networks and their Applications, DCNA 2021 - Kaliningrad, Russian Federation
Duration: 13 Sep 202115 Sep 2021
http://bfnaics.kantiana.ru/

Publication series

NameConference Proceedings - 5th Scientific School Dynamics of Complex Networks and their Applications, DCNA 2021

Conference

Conference5th Scientific School on Dynamics of Complex Networks and their Applications, DCNA 2021
Abbreviated titleDCNA 2021
Country/TerritoryRussian Federation
CityKaliningrad
Period13/09/2115/09/21
Internet address

Scopus subject areas

  • Artificial Intelligence
  • Computer Networks and Communications
  • Computer Science Applications
  • Control and Optimization

Keywords

  • Consensus
  • parameter estimation
  • SPSA

Fingerprint

Dive into the research topics of 'Distributed Stochastic Optimization with Heavy-Ball Momentum Term for Parameter Estimation'. Together they form a unique fingerprint.

Cite this