Multi-class Generative Adversarial Networks with the L2 Loss Function.

Generative adversarial networks (GANs) have achieved huge success in unsupervised learning. Most of GANs treat the discriminator as a classifier with the binary sigmoid cross entropy loss function. However, we find that the sigmoid cross entropy loss function will sometimes lead to the saturation problem in GANs learning. In this work, we propose to adopt the L2 loss function for the discriminator. The properties of the L2 loss function can improve the stabilization of GANs learning. With the usage of the L2 loss function, we propose the multi-class generative adversarial networks for the purpose of image generation with multiple classes. We evaluate the multi-class GANs on a handwritten Chinese characters dataset with 3740 classes. The experiments demonstrate that the multi-class GANs can generate elegant images on datasets with a large number of classes. Comparison experiments between the L2 loss function and the sigmoid cross entropy loss function are also conducted and the results demonstrate the stabilization of the L2 loss function.