SVGD: A Virtual Gradients Descent Method for Stochastic Optimization
arXiv:1907.04021
The paper introduces Stochastic Virtual Gradient Descent (SVGD), a memory‑efficient stochastic optimization algorithm that defines gradients via computational graphs and automatic differentiation, and provides convergence analysis and experimental validation.
Abstract
Inspired by dynamic programming, we propose Stochastic Virtual Gradient Descent (SVGD) algorithm where the Virtual Gradient is defined by computational graph and automatic differentiation. The method is computationally efficient and has little memory requirements. We also analyze the theoretical convergence properties and implementation of the algorithm. Experimental results on multiple datasets and network models show that SVGD has advantages over other stochastic optimization methods.
12 pages, 12 figures, conference papers