Stochastic Gradient Push for Distributed Deep Learning

TL;DR

This paper introduces Stochastic Gradient Push (SGP), a distributed deep learning algorithm combining PushSum gossip with stochastic gradients, which converges reliably and is robust to communication delays, improving training efficiency.

Contribution

It presents SGP, a novel distributed training method that ensures convergence and consensus despite communication challenges, extending PushSum gossip with stochastic gradients.

Findings

SGP converges to a stationary point at the same rate as SGD.

All nodes reach consensus during training.

Empirical results show SGP's effectiveness on ImageNet and translation tasks.

Abstract

Distributed data-parallel algorithms aim to accelerate the training of deep neural networks by parallelizing the computation of large mini-batch gradient updates across multiple nodes. Approaches that synchronize nodes using exact distributed averaging (e.g., via AllReduce) are sensitive to stragglers and communication delays. The PushSum gossip algorithm is robust to these issues, but only performs approximate distributed averaging. This paper studies Stochastic Gradient Push (SGP), which combines PushSum with stochastic gradient updates. We prove that SGP converges to a stationary point of smooth, non-convex objectives at the same sub-linear rate as SGD, and that all nodes achieve consensus. We empirically validate the performance of SGP on image classification (ResNet-50, ImageNet) and machine translation (Transformer, WMT'16 En-De) workloads. Our code will be made publicly available.