Distributed Stochastic Gradient Descent with Staleness: A Stochastic Delay Differential Equation Based Framework

TL;DR

This paper introduces a stochastic delay differential equation framework to analyze and optimize asynchronous distributed SGD, accounting for delays and staleness, and revealing conditions for convergence and acceleration.

Contribution

The paper develops a unified SDDE-based framework for analyzing and optimizing asynchronous SGD with delays, without assuming memoryless computation times, and provides new insights into scheduling policies.

Findings

Increasing workers does not always speed up SGD due to staleness.

Small staleness may not slow convergence, large staleness can cause divergence.

The framework effectively models complex non-convex learning tasks.

Abstract

Distributed stochastic gradient descent (SGD) has attracted considerable recent attention due to its potential for scaling computational resources, reducing training time, and helping protect user privacy in machine learning. However, the staggers and limited bandwidth may induce random computational/communication delays, thereby severely hindering the learning process. Therefore, how to accelerate asynchronous SGD by efficiently scheduling multiple workers is an important issue. In this paper, a unified framework is presented to analyze and optimize the convergence of asynchronous SGD based on stochastic delay differential equations (SDDEs) and the Poisson approximation of aggregated gradient arrivals. In particular, we present the run time and staleness of distributed SGD without a memorylessness assumption on the computation times. Given the learning rate, we reveal the relevant SDDE's damping coefficient and its delay statistics, as functions of the number of activated clients, staleness threshold, the eigenvalues of the Hessian matrix of the objective function, and the overall computational/communication delay. The formulated SDDE allows us to present both the distributed SGD's convergence condition and speed by calculating its characteristic roots, thereby optimizing the scheduling policies for asynchronous/event-triggered SGD. It is interestingly shown that increasing the number of activated workers does not necessarily accelerate distributed SGD due to staleness. Moreover, a small degree of staleness does not necessarily slow down the convergence, while a large degree of staleness will result in the divergence of distributed SGD. Numerical results demonstrate the potential of our SDDE framework, even in complex learning tasks with non-convex objective functions.