Distributed Coordinate Descent for L1-regularized Logistic Regression

TL;DR

This paper introduces d-GLMNET, a distributed algorithm for L1-regularized logistic regression, demonstrating its effectiveness on large datasets that cannot be processed on a single machine.

Contribution

The paper presents a novel distributed coordinate descent algorithm specifically designed for L1-regularized logistic regression, outperforming existing online learning methods.

Findings

d-GLMNET outperforms distributed online learning methods

Effective for very large datasets

Demonstrates superior convergence and accuracy

Abstract

Solving logistic regression with L1-regularization in distributed settings is an important problem. This problem arises when training dataset is very large and cannot fit the memory of a single machine. We present d-GLMNET, a new algorithm solving logistic regression with L1-regularization in the distributed settings. We empirically show that it is superior over distributed online learning via truncated gradient.