Constrained Generative Adversarial Networks
Generative Adversarial Networks (GANs) are a powerful subclass of generative models. Yet, how to effectively train them to reach Nash equilibrium is a challenge. A number of experiments have indicated that one possible solution is to bound the function space of the discriminator. In practice, when o...
Main Authors: | , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
IEEE
2021-01-01
|
Series: | IEEE Access |
Subjects: | |
Online Access: | https://ieeexplore.ieee.org/document/9335934/ |
id |
doaj-b1d4db230d804e86ae0a365f163a2562 |
---|---|
record_format |
Article |
spelling |
doaj-b1d4db230d804e86ae0a365f163a25622021-03-30T15:17:34ZengIEEEIEEE Access2169-35362021-01-019192081921810.1109/ACCESS.2021.30548229335934Constrained Generative Adversarial NetworksXiaopeng Chao0https://orcid.org/0000-0002-7494-8668Jiangzhong Cao1Yuqin Lu2Qingyun Dai3Shangsong Liang4https://orcid.org/0000-0003-1625-2168School of Information Engineering, Guangdong University of Technology, Guangzhou, ChinaSchool of Information Engineering, Guangdong University of Technology, Guangzhou, ChinaSchool of Information Engineering, Guangdong University of Technology, Guangzhou, ChinaSchool of Electronic and Information Engineering, Guangdong Polytechnic Normal University, Guangzhou, ChinaSchool of Computer Science and Technology, Sun Yat-sen University, Guangzhou, ChinaGenerative Adversarial Networks (GANs) are a powerful subclass of generative models. Yet, how to effectively train them to reach Nash equilibrium is a challenge. A number of experiments have indicated that one possible solution is to bound the function space of the discriminator. In practice, when optimizing the standard loss function without limiting the discriminator's output, the discriminator may suffer from lack of convergence. To be able to reach the Nash equilibrium in a faster way during training and obtain better generative data, we propose constrained generative adversarial networks, GAN-C, where a constraint on the discriminator's output is introduced. We theoretically prove that our proposed loss function shares the same Nash equilibrium as the standard one, and our experiments on mixture of Gaussians, MNIST, CIFAR-10, STL-10, FFHQ, and CAT datasets show that our loss function can better stabilize training and yield even better high-quality images.https://ieeexplore.ieee.org/document/9335934/Generative adversarial networksNash equilibriumLipschitz constraint |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Xiaopeng Chao Jiangzhong Cao Yuqin Lu Qingyun Dai Shangsong Liang |
spellingShingle |
Xiaopeng Chao Jiangzhong Cao Yuqin Lu Qingyun Dai Shangsong Liang Constrained Generative Adversarial Networks IEEE Access Generative adversarial networks Nash equilibrium Lipschitz constraint |
author_facet |
Xiaopeng Chao Jiangzhong Cao Yuqin Lu Qingyun Dai Shangsong Liang |
author_sort |
Xiaopeng Chao |
title |
Constrained Generative Adversarial Networks |
title_short |
Constrained Generative Adversarial Networks |
title_full |
Constrained Generative Adversarial Networks |
title_fullStr |
Constrained Generative Adversarial Networks |
title_full_unstemmed |
Constrained Generative Adversarial Networks |
title_sort |
constrained generative adversarial networks |
publisher |
IEEE |
series |
IEEE Access |
issn |
2169-3536 |
publishDate |
2021-01-01 |
description |
Generative Adversarial Networks (GANs) are a powerful subclass of generative models. Yet, how to effectively train them to reach Nash equilibrium is a challenge. A number of experiments have indicated that one possible solution is to bound the function space of the discriminator. In practice, when optimizing the standard loss function without limiting the discriminator's output, the discriminator may suffer from lack of convergence. To be able to reach the Nash equilibrium in a faster way during training and obtain better generative data, we propose constrained generative adversarial networks, GAN-C, where a constraint on the discriminator's output is introduced. We theoretically prove that our proposed loss function shares the same Nash equilibrium as the standard one, and our experiments on mixture of Gaussians, MNIST, CIFAR-10, STL-10, FFHQ, and CAT datasets show that our loss function can better stabilize training and yield even better high-quality images. |
topic |
Generative adversarial networks Nash equilibrium Lipschitz constraint |
url |
https://ieeexplore.ieee.org/document/9335934/ |
work_keys_str_mv |
AT xiaopengchao constrainedgenerativeadversarialnetworks AT jiangzhongcao constrainedgenerativeadversarialnetworks AT yuqinlu constrainedgenerativeadversarialnetworks AT qingyundai constrainedgenerativeadversarialnetworks AT shangsongliang constrainedgenerativeadversarialnetworks |
_version_ |
1724179701791981568 |