Logo image
Back
Distributed Dual Coordinate Ascent in a Ring Network for Distributed Machine Learning Process
Conference proceeding

Distributed Dual Coordinate Ascent in a Ring Network for Distributed Machine Learning Process

Myung Cho, Sara Ali, Marjan Asadinia, Lifeng Lai and Weiyu Xu
Conference record - Asilomar Conference on Signals, Systems, & Computers, pp.1806-1811
10/26/2025
DOI: 10.1109/IEEECONF67917.2025.11443696

View Online

Abstract

This paper investigates a distributed machine learning process in which datasets are stored across nodes in a ring network. The learning task involves solving a regularized loss minimization problem using the distributed data. Assuming the network topology is fixed and only the communication frequency between nodes can be adjusted, we examine two network structures: the original ring network and a spanning tree derived from it. Our focus is to study the impact of network topology - what we refer to as the network topology effect - on the performance of Distributed Dual Coordinate Ascent (DDCA) in convergence speed. Through simulation experiments on the Wine Quality regression and MNIST classification tasks, we demonstrate that the ring network yields faster convergence behavior than the corresponding tree network in the context of distributed dual coordinate ascent.
 Machine Learning Computers Convergence Distributed databases distributed dataset distributed dual coordinate ascent distributed machine learning machine learning over communication networks Minimization Network topology Regression tree analysis ring network Spanning trees Structural rings Trees (botanical)

Details

Metrics

1 Record Views
Logo image