Proximal Stochastic Dual Coordinate Ascent

TL;DR

This paper introduces a proximal dual coordinate ascent algorithm that effectively handles regularized loss minimization problems, achieving competitive convergence rates and broad applicability.

Contribution

It presents a novel proximal dual coordinate ascent framework applicable to various regularized problems, improving convergence rates over existing methods.

Findings

Achieves convergence rates matching or exceeding state-of-the-art.

Applicable to $\,ell_1$ regularization and structured output SVM.

Demonstrates broad utility across multiple regularized loss minimization tasks.

Abstract

We introduce a proximal version of dual coordinate ascent method. We demonstrate how the derived algorithmic framework can be used for numerous regularized loss minimization problems, including $\ell_1$ regularization and structured output SVM. The convergence rates we obtain match, and sometimes improve, state-of-the-art results.