Improve Training Stability of Semi-supervised Generative Adversarial Networks with Collaborative Training

Abstract

Improved generative adversarial network (Improved GAN) is a successful method of using generative adversarial models to solve the problem of semi-supervised learning. However, it suffers from the problem of unstable training. In this paper, we found that the instability is mostly due to the vanishing gradients on the generator. To remedy this issue, we propose a new method to use collaborative training to improve the stability of semi-supervised GAN with the combination of Wasserstein GAN. The experiments have shown that our proposed method is more stable than the original Improved GAN and achieves comparable classification accuracy on different data sets.

Tasks

Reproductions