Adversarial Autoencoder training procedure does not correspond to procedure described in paper

[acse-zrw20]

The aae.py file trains the adversarial autoencoder in a different way to how it is described in the paper (https://arxiv.org/pdf/1511.05644.pdf).

In the paper there is a reconstruction phase, where the autoencoder minimizes the reconstruction error. And a regularization phase, where the adversarial network first updates its discriminative network and then the generator to fool the discriminative network.

In your code, the discriminator is first trained based on real and fake latent vectors, and then the generator and autoencoder are trained simultaneously with respective weights to the losses.