# DL autoencoder * Regularization * Denoising * Undercomplete Overcomplete * Overcomplete case in which the hidden code hasdimension greater than the input * An autoencoder whose code dimension is lessthan the input dimension is calledundercomplete * Contractive autoencoder Penalize 'unwanted variations' Frobenius norm of Jacobian * Variational Autoencoder * How to generate samples from an auto encoder: * Make hidden layer a distribution which is easy to sample from # Generative Adversarial Networks * Generate network * Discriminator network * Why is difficult to optimize * How to evluate generative model