Asynchronous ADMM via a data exchange server
With advances in inter-processing-unit communication technology, distributed algorithms are becoming increasingly advantageous. This paper focuses on solving convex distributed optimisation problems with local consensus coupling constraints via the alternating direction method of multipliers (ADMM),...
| Auteurs principaux: | , |
|---|---|
| Format: | Journal article |
| Langue: | English |
| Publié: |
IEEE
2024
|
| Résumé: | With advances in inter-processing-unit communication technology, distributed algorithms are becoming increasingly advantageous. This paper focuses on solving convex distributed optimisation problems with local consensus coupling constraints via the alternating direction method of multipliers (ADMM), by means of an asynchronous methodology allowing for communication delays. We use a bipartite undirected graph to denote the update structure of the processing agents that cooperatively perform the distributed algorithm without a centralised aggregator. We introduce a data server to exchange the asynchronous consensus data among the processing agents. Under certain technical assumptions that involve bounded delays, bounded step sizes, and strong convexities in parts of the local objectives, the running average of the local iterates generated by the proposed asynchronous algorithm converge to an optimal solution. |
|---|