Communication-Efficient Distributed Learning via Lazily Aggregated Quantized Gradients

TL;DR

This paper introduces LAQ, a communication-efficient distributed learning method that combines gradient quantization and selective communication skipping, achieving convergence comparable to traditional methods with reduced communication costs.

Contribution

The paper proposes LAQ, a novel approach that adaptively compresses and selectively skips gradient communications, significantly reducing communication overhead while maintaining convergence rates.

Findings

LAQ achieves the same linear convergence rate as gradient descent in strongly convex settings.

LAQ reduces communication bits and rounds compared to existing algorithms.

Empirical results confirm significant communication savings with real data.

Abstract

The present paper develops a novel aggregated gradient approach for distributed machine learning that adaptively compresses the gradient communication. The key idea is to first quantize the computed gradients, and then skip less informative quantized gradient communications by reusing outdated gradients. Quantizing and skipping result in `lazy' worker-server communications, which justifies the term Lazily Aggregated Quantized gradient that is henceforth abbreviated as LAQ. Our LAQ can provably attain the same linear convergence rate as the gradient descent in the strongly convex case, while effecting major savings in the communication overhead both in transmitted bits as well as in communication rounds. Empirically, experiments with real data corroborate a significant communication reduction compared to existing gradient- and stochastic gradient-based algorithms.