Communication-Efficient Distributed Dual Coordinate Ascent

TL;DR

This paper introduces CoCoA, a communication-efficient distributed optimization framework that leverages local computation to significantly reduce communication costs, achieving faster convergence in large-scale machine learning tasks.

Contribution

The paper presents a novel primal-dual framework, CoCoA, with proven convergence rates and practical implementation, outperforming existing mini-batch algorithms in speed.

Findings

CoCoA converges 25 times faster than state-of-the-art methods.

Achieves the same solution quality with less communication.

Demonstrated effectiveness on real-world datasets.

Abstract

Communication remains the most significant bottleneck in the performance of distributed optimization algorithms for large-scale machine learning. In this paper, we propose a communication-efficient framework, CoCoA, that uses local computation in a primal-dual setting to dramatically reduce the amount of necessary communication. We provide a strong convergence rate analysis for this class of algorithms, as well as experiments on real-world distributed datasets with implementations in Spark. In our experiments, we find that as compared to state-of-the-art mini-batch versions of SGD and SDCA algorithms, CoCoA converges to the same .001-accurate solution quality on average 25x as quickly.