Teaching a GAN What Not to Learn

Neural Information Processing Systems 

Generative adversarial networks (GANs) were originally envisioned as unsupervised generative models that learn to follow a target distribution.