BourGAN: Generative Networks with Metric Embeddings

Neural Information Processing Systems 

This paper addresses the mode collapse for generative adversarial networks (GANs). We view modes as a geometric structure of data distribution in a metric space.