HOGWILD!: A Lock-Free Approach to Parallelizing Stochastic Gradient Descent

TLDR

Stochastic Gradient Descent is widely used for high‑performance machine learning, yet existing parallelization methods rely on costly memory locking. The authors demonstrate that SGD can be executed without locks by developing a novel theoretical framework, algorithm, and implementation. They introduce HOGWILD!, an update scheme that permits processors to write to shared memory concurrently, and experimentally validate its effectiveness. For sparse optimization problems, HOGWILD!

Abstract

Stochastic Gradient Descent (SGD) is a popular algorithm that can achieve state-of-the-art performance on a variety of machine learning tasks. Several researchers have recently proposed schemes to parallelize SGD, but all require performance-destroying memory locking and synchronization. This work aims to show using novel theoretical analysis, algorithms, and implementation that SGD can be implemented without any locking. We present an update scheme called HOGWILD! which allows processors access to shared memory with the possibility of overwriting each other's work. We show that when the associated optimization problem is sparse, meaning most gradient updates only modify small parts of the decision variable, then HOGWILD! achieves a nearly optimal rate of convergence. We demonstrate experimentally that HOGWILD! outperforms alternative schemes that use locking by an order of magnitude.