ICML 2020poster47 citations

SGD Learns One-Layer Networks in WGANs

Qi Lei, Jason Lee, Alex Dimakis, Constantinos Daskalakis

Abstract

Generative adversarial networks (GANs) are a widely used framework for learning generative models. Wasserstein GANs (WGANs), one of the most successful variants of GANs, require solving a minmax optimization problem to global optimality, but are in practice successfully trained using stochastic gradient descent-ascent. In this paper, we show that, when the generator is a one-layer network, stochastic gradient descent-ascent converges to a global solution with polynomial time and sample complexity.

BibTeX
@InProceedings{pmlr-v119-lei20b,
  title = 	 {{SGD} Learns One-Layer Networks in {WGAN}s},
  author =       {Lei, Qi and Lee, Jason and Dimakis, Alex and Daskalakis, Constantinos},
  booktitle = 	 {Proceedings of the 37th International Conference on Machine Learning},
  pages = 	 {5799--5808},
  year = 	 {2020},
  editor = 	 {III, Hal Daumé and Singh, Aarti},
  volume = 	 {119},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {13--18 Jul},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v119/lei20b/lei20b.pdf},
  url = 	 {https://proceedings.mlr.press/v119/lei20b.html},
  abstract = 	 {Generative adversarial networks (GANs) are a widely used framework for learning generative models. Wasserstein GANs (WGANs), one of the most successful variants of GANs, require solving a minmax optimization problem to global optimality, but are in practice successfully trained using stochastic gradient descent-ascent. In this paper, we show that, when the generator is a one-layer network, stochastic gradient descent-ascent converges to a global solution with polynomial time and sample complexity.}
}
SGD Learns One-Layer Networks in WGANs · ICML 2020