Staleness-aware Async-SGD for Distributed Deep Learning

TL;DR

This paper introduces a staleness-aware variant of asynchronous SGD for distributed deep learning, which adaptively adjusts the learning rate based on gradient staleness to improve convergence and training speed.

Contribution

The paper proposes a novel ASGD algorithm that modulates learning rate according to gradient staleness, with theoretical convergence guarantees and improved empirical performance.

Findings

Outperforms standard ASGD and SSGD on CIFAR10 and ImageNet benchmarks.

Provides theoretical convergence guarantees for the proposed algorithm.

Demonstrates faster and more stable training in distributed deep learning settings.

Abstract

Deep neural networks have been shown to achieve state-of-the-art performance in several machine learning tasks. Stochastic Gradient Descent (SGD) is the preferred optimization algorithm for training these networks and asynchronous SGD (ASGD) has been widely adopted for accelerating the training of large-scale deep networks in a distributed computing environment. However, in practice it is quite challenging to tune the training hyperparameters (such as learning rate) when using ASGD so as achieve convergence and linear speedup, since the stability of the optimization algorithm is strongly influenced by the asynchronous nature of parameter updates. In this paper, we propose a variant of the ASGD algorithm in which the learning rate is modulated according to the gradient staleness and provide theoretical guarantees for convergence of this algorithm. Experimental verification is performed on commonly-used image classification benchmarks: CIFAR10 and Imagenet to demonstrate the superior effectiveness of the proposed approach, compared to SSGD (Synchronous SGD) and the conventional ASGD algorithm.