Generative Adversarial Networks with Joint Distribution Moment Matching
- PDF / 1,786,186 Bytes
- 19 Pages / 439.37 x 666.142 pts Page_size
- 104 Downloads / 258 Views
Generative Adversarial Networks with Joint Distribution Moment Matching Yi-Ying Zhang1 · Chao-Min Shen1,2,3 · Hao Feng4 · Preston Thomas Fletcher5 · Gui-Xu Zhang1 Received: 2 November 2018 / Revised: 27 March 2019 / Accepted: 1 April 2019 © Operations Research Society of China, Periodicals Agency of Shanghai University, Science Press, and Springer-Verlag GmbH Germany, part of Springer Nature 2019
Abstract Generative adversarial networks (GANs) have shown impressive power in the field of machine learning. Traditional GANs have focused on unsupervised learning tasks. In recent years, conditional GANs that can generate data with labels have been proposed in semi-supervised learning and have achieved better image quality than traditional GANs. Conditional GANs, however, generally only minimize the difference between marginal distributions of real and generated data, neglecting the difference with respect to each class of the data. To address this challenge, we propose the GAN with joint dis-
This work is supported by the National Natural Science Foundation of China (Nos. 11771276, 11471208, 61731009), and the Foundation of Science and Technology Commission of Shanghai Municipality (No. 14DZ2260800).
B
Chao-Min Shen [email protected] Yi-Ying Zhang [email protected] Hao Feng [email protected] Preston Thomas Fletcher [email protected] Gui-Xu Zhang [email protected]
1
Department of Computer Science, East China Normal University, Shanghai 200062, China
2
Shanghai Key Laboratory of Multidimensional Information Processing, East China Normal University, Shanghai 200062, China
3
Westlake Institute for Brain-Like Science and Technology, Hangzhou 310027, China
4
Didi Chuxing Science and Technology Co. Ltd., Beijing 100193, China
5
Department of Electrical and Computer Engineering, University of Virginia, Charlottesville, VA 22904, USA
123
Y. Zhang et al.
tribution moment matching (JDMM-GAN) for matching the joint distribution based on maximum mean discrepancy, which minimizes the differences of both the marginal and conditional distributions. The learning procedure is iteratively conducted by the stochastic gradient descent and back-propagation. We evaluate JDMM-GAN on several benchmark datasets, including MNIST, CIFAR-10 and the Extended Yale Face. Compared with the state-of-the-art GANs, JDMM-GAN generates more realistic images and achieves the best inception score for CIFAR-10 dataset. Keywords Generative Adversarial Networks · Joint Distribution Moment Matching · Maximum mean discrepancy Mathematics Subject Classification 68T45 · 62P99
1 Introduction Generative adversarial networks (GANs) [1] have become a prevailing topic in machine learning since Goodfellow et al. proposed them in 2014. Their motivation is simple yet effective: With samples from the real data distribution, a generator network, G, is trained to synthesize random noise into generated data. Meanwhile, another adversarial discriminator, D, is trained to distinguish the real data from the generated data. The training procedure
Data Loading...
