Optimal convergence rates of totally asynchronous optimization

Asynchronous optimization algorithms are at the core of modern machine learning and resource allocation systems. However, most convergence results consider bounded information delays and several important algorithms lack guarantees when they operate under total asynchrony. In this paper, we derive e...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Proceedings of the IEEE Conference on Decision & Control s. 6484 - 6490
Hlavní autoři: Wu, Xuyang, Magnusson, Sindri, Reza Feyzmahdavian, Hamid, Johansson, Mikael
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 06.12.2022
Témata:
Computational modeling
Delays
Indexes
Machine learning
Machine learning algorithms
Numerical models
Resource management
ISSN:2576-2370
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Asynchronous optimization algorithms are at the core of modern machine learning and resource allocation systems. However, most convergence results consider bounded information delays and several important algorithms lack guarantees when they operate under total asynchrony. In this paper, we derive explicit convergence rates for the proximal incremental aggregated gradient (PIAG) and the asynchronous block-coordinate descent (Async-BCD) methods under a specific model of total asynchrony, and show that the derived rates are order-optimal. The convergence bounds provide an insightful understanding of how the growth rate of the delays deteriorates the convergence times of the algorithms. Our theoretical findings are demonstrated by a numerical example.
ISSN:2576-2370
DOI:10.1109/CDC51059.2022.9993168