SVGD: A Virtual Gradients Descent Method for Stochastic Optimization
Machine Learning
2019-08-01 v2 Optimization and Control
Machine Learning
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.
Cite
@article{arxiv.1907.04021,
title = {SVGD: A Virtual Gradients Descent Method for Stochastic Optimization},
author = {Zheng Li and Shi Shu},
journal= {arXiv preprint arXiv:1907.04021},
year = {2019}
}
Comments
12 pages, 12 figures, conference papers