Gradient Sparification for Asynchronous Distributed Training

TL;DR

This paper introduces a dual-way gradient sparsification method for asynchronous distributed training, reducing communication costs and improving scalability in federated learning scenarios.

Contribution

The paper proposes a novel dual-way gradient sparsification technique and a sparsification aware momentum to enable efficient asynchronous training with reduced communication overhead.

Findings

Achieves lower communication cost with the same compression ratio.

Improves scalability and generalization in asynchronous distributed training.

Demonstrates effectiveness on a 32-worker cluster.

Abstract

Modern large scale machine learning applications require stochastic optimization algorithms to be implemented on distributed computational architectures. A key bottleneck is the communication overhead for exchanging information, such as stochastic gradients, among different nodes. Recently, gradient sparsification techniques have been proposed to reduce communications cost and thus alleviate the network overhead. However, most of gradient sparsification techniques consider only synchronous parallelism and cannot be applied in asynchronous scenarios, such as asynchronous distributed training for federated learning at mobile devices. In this paper, we present a dual-way gradient sparsification approach (DGS) that is suitable for asynchronous distributed training. We let workers download model difference, instead of the global model, from the server, and the model difference information is also sparsified so that the information exchanged overhead is reduced by sparsifying the dual-way communication between the server and workers. To preserve accuracy under dual-way sparsification, we design a sparsification aware momentum (SAMomentum) to turn sparsification into adaptive batch size between each parameter. We conduct experiments at a cluster of 32 workers, and the results show that, with the same compression ratio but much lower communication cost, our approach can achieve better scalability and generalization ability.