Preconditioning ADMM for Fast Decentralized Optimization

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this work, we consider the distributed optimization problem using networked computing machines. Specifically, we are interested in solving this problem using the alternating direction method of multipliers (ADMM) while accounting for edge weights. Existing works focus on star graphs and use simple heuristics for other types of graphs. The present work shows that the optimal edge weights design is equivalent to the preconditioning matrix of ADMM that leads to the fastest convergence speed. Based on a tight convergence rate of ADMM, we show that the preconditioning matrix of general graphs can be found by minimizing the ratio of the largest and smallest nonzero eigenvalue of the graph Laplacian. Numerical experiments show that preconditioned ADMM converges much faster to a certain accuracy with less communication rounds, and exemplify the robustness to topology changes of the underlying network.

Original languageEnglish (US)
Title of host publication2020 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2020 - Proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages3142-3146
Number of pages5
ISBN (Electronic)9781509066315
DOIs
StatePublished - May 2020
Externally publishedYes
Event2020 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2020 - Barcelona, Spain
Duration: May 4 2020May 8 2020

Publication series

NameICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
Volume2020-May
ISSN (Print)1520-6149

Conference

Conference2020 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2020
CountrySpain
CityBarcelona
Period5/4/205/8/20

Bibliographical note

Funding Information:
This work is supported by NSF grant 1901134.

Keywords

  • ADMM
  • Decentralized optimization
  • hybrid ADMM
  • preconditioning

Fingerprint Dive into the research topics of 'Preconditioning ADMM for Fast Decentralized Optimization'. Together they form a unique fingerprint.

Cite this