Have two networks: a generative network $G(z)$, which tries to generate images which look like true images, starting from a vector of random numbers$z$, and a discriminator network,$D(Y)$.

## Background

### Mode collapse

A commonly encountered failure case for GANs where the generator learns to produce samples with extremely low variety.

## Example architectures

#### DCGAN

• https://github.com/Newmu/dcgan_code (theano)
• https://github.com/carpedm20/DCGAN-tensorflow

#### WassersteinGAN

• https://github.com/martinarjovsky/WassersteinGAN