The ADMM algorithm for distributed averaging: convergence rates and optimal parameter selection
2014 (English)In: 48th Asilomar Conference on Signals, Systems, and Computers 2014, IEEE conference proceedings, 2014, 783-787 p.Conference paper (Refereed)
We derive the optimal step-size and overrelaxationparameter that minimizes the convergence time oftwo ADMM-based algorithms for distributed averaging. Ourstudy shows that the convergence times for given step-size andover-relaxation parameters depend on the spectral propertiesof the normalized Laplacian of the underlying communicationgraph. Motivated by this, we optimize the edge-weights of thecommunication graph to improve the convergence speed evenfurther. The performance of the ADMM algorithms with ourparameter selection are compared with alternatives from theliterature in extensive numerical simulations on random graphs.
Place, publisher, year, edition, pages
IEEE conference proceedings, 2014. 783-787 p.
IdentifiersURN: urn:nbn:se:kth:diva-159704DOI: 10.1109/ACSSC.2014.7094556ISBN: 978-1-4799-8295-0OAI: oai:DiVA.org:kth-159704DiVA: diva2:787052
48th Asilomar Conference on Signals, Systems, and Computers, November 2-5 2014,Pacific Grove, CA, USA
QC 201505112015-02-092015-02-092015-05-11Bibliographically approved