EUROPT 2024
Abstract Submission

271. First-order splitting methods for decentralized optimization

Invited abstract in session WF-3: Splitting algorithms, stream Variational analysis: theory and algorithms.

Wednesday, 16:20 - 18:00
Room: M:J

Authors (first author is the speaker)

1. Felipe Atenas
University of Melbourne
2. Matthew Tam
School of Mathematics and Statistics, University of Melbourne
3. Minh N. Dao
RMIT University

Abstract

We consider finitely many agents over a connected network, whose aim is to cooperatively solve a convex optimization problem with composite sum structure, without having direct access to the information of the overall network. We propose primal-dual splitting methods of proximal-gradient type that use no central coordinator, performing forward-backward steps separately for each agent, and rounds of communication among neighbors. The algorithms we propose extend the so-called PG-EXTRA method for decentralized minimization problems, to three-operator splitting and convex-concave min-max problems.

Keywords

Status: accepted


Back to the list of papers