Variance Reduced Stochastic Gradient Descent with Neighbors

Abstract

Stochastic Gradient Descent (SGD) is a workhorse in machine learning, yet its slow convergence can be a computational bottleneck. Variance reduction techniques such as SAG, SVRG and SAGA have been proposed to overcome this weakness, achieving linear convergence. However, these methods are either based on computations of full gradients at pivot points, or on keeping per data point corrections in memory. Therefore speed-ups relative to SGD may need a minimal number of epochs in order to materialize. This paper investigates algorithms that can exploit neighborhood structure in the training data to share and re-use information about past stochastic gradients across data points, which offers advantages in the transient optimization phase. As a side-product we provide a unified convergence analysis for a family of variance reduction algorithms, which we call memorization algorithms. We provide experimental results supporting our theory.

Appears in: Advances in Neural Information Processing Systems 28 (NIPS 2015). 13 pages

References in corpus (1)

• Optimal Data-Dependent Hashing for Approximate Near Neighbors

Cited by in corpus (9)

• A Simple Practical Accelerated Method for Finite Sums
• Improved asynchronous parallel optimization analysis for stochastic incremental methods
• On the Theory of Variance Reduction for Stochastic Gradient Monte Carlo
• A Unified Analysis of Stochastic Gradient Methods for Nonconvex Federated Optimization
• SMART: The Stochastic Monotone Aggregated Root-Finding Algorithm
• DynaNewton - Accelerating Newton's Method for Machine Learning
• Variance-Reduced Proximal Stochastic Gradient Descent for Non-convex Composite optimization
• Lightweight Stochastic Optimization for Minimizing Finite Sums with Infinite Data
• Stochastic Polyak Stepsize with a Moving Target