Reproduces Adversarial Autoencoder architecture from Makhzani, Alireza, et al. "Adversarial autoencoders." arXiv preprint arXiv:1511.05644 (2015) with Keras.

The Adversarial Autoencoder behaves similarly to Variational Autoencoders, forcing the latent space of an autoencoder to follow a predefined prior. In the case of the Adversarial Autoencoder, this latent space can be defined arbitrarily and easily sampled and fed into the Discriminator in the network.

The left image shows the latent space of an unseen MNIST test set after training with an Adversarial Autoencoder for 50 epochs, which follows a 2D Gaussian prior. Contrast this with the latent space of the regular Autoencoder trained under the same conditions, with a far more irregular latent distribution.

To train a model just run

$ python keras-aae.py --train

For more parameters, run with --help flag.

For comparison with a regular autoencoder, run

$ python regular-ae.py --train --noadversarial