Accelerated Mini-Batch Stochastic Dual Coordinate Ascent

TL;DR

This paper introduces an accelerated mini-batch stochastic dual coordinate ascent method that improves convergence rates for regularized loss minimization problems in machine learning, with practical parallel implementation.

Contribution

It presents a novel accelerated mini-batch SDCA algorithm with proven fast convergence, extending SDCA to more efficient parallelizable optimization.

Findings

Proves a faster convergence rate for the proposed method.

Demonstrates improved performance over vanilla SDCA.

Shows competitiveness with accelerated gradient descent methods.

Abstract

Stochastic dual coordinate ascent (SDCA) is an effective technique for solving regularized loss minimization problems in machine learning. This paper considers an extension of SDCA under the mini-batch setting that is often used in practice. Our main contribution is to introduce an accelerated mini-batch version of SDCA and prove a fast convergence rate for this method. We discuss an implementation of our method over a parallel computing system, and compare the results to both the vanilla stochastic dual coordinate ascent and to the accelerated deterministic gradient descent method of \cite{nesterov2007gradient}.