We study the effect of the stochastic gradient noise on the training of
generative adversarial networks (GANs) and show that it can prevent the
convergence of standard game optimization methods, while the batch version
converges. We address this issue with a novel stochastic variance-reduced
extragradient (SVRE) optimization algorithm that improves upon the best
convergence rates proposed in the literature. We observe empirically that SVRE
performs similarly to a batch method on MNIST while being computationally
cheaper, and that SVRE yields more stable GAN training on standard datasets.
Описание
[1904.08598] Reducing Noise in GAN Training with Variance Reduced Extragradient
%0 Journal Article
%1 chavdarova2019reducing
%A Chavdarova, Tatjana
%A Gidel, Gauthier
%A Fleuret, François
%A Lacoste-Julien, Simon
%D 2019
%K adversarial generative-models optimization
%T Reducing Noise in GAN Training with Variance Reduced Extragradient
%U http://arxiv.org/abs/1904.08598
%X We study the effect of the stochastic gradient noise on the training of
generative adversarial networks (GANs) and show that it can prevent the
convergence of standard game optimization methods, while the batch version
converges. We address this issue with a novel stochastic variance-reduced
extragradient (SVRE) optimization algorithm that improves upon the best
convergence rates proposed in the literature. We observe empirically that SVRE
performs similarly to a batch method on MNIST while being computationally
cheaper, and that SVRE yields more stable GAN training on standard datasets.
@article{chavdarova2019reducing,
abstract = {We study the effect of the stochastic gradient noise on the training of
generative adversarial networks (GANs) and show that it can prevent the
convergence of standard game optimization methods, while the batch version
converges. We address this issue with a novel stochastic variance-reduced
extragradient (SVRE) optimization algorithm that improves upon the best
convergence rates proposed in the literature. We observe empirically that SVRE
performs similarly to a batch method on MNIST while being computationally
cheaper, and that SVRE yields more stable GAN training on standard datasets.},
added-at = {2019-09-04T15:51:55.000+0200},
author = {Chavdarova, Tatjana and Gidel, Gauthier and Fleuret, François and Lacoste-Julien, Simon},
biburl = {https://www.bibsonomy.org/bibtex/2940c9d7f84bc89ba3584f72104075cb9/kirk86},
description = {[1904.08598] Reducing Noise in GAN Training with Variance Reduced Extragradient},
interhash = {b92ac8ee05b811f4aca72f07bb63664c},
intrahash = {940c9d7f84bc89ba3584f72104075cb9},
keywords = {adversarial generative-models optimization},
note = {cite arxiv:1904.08598},
timestamp = {2019-09-04T15:51:55.000+0200},
title = {Reducing Noise in GAN Training with Variance Reduced Extragradient},
url = {http://arxiv.org/abs/1904.08598},
year = 2019
}