An autoencoder that differs from others (with deterministic encoder) by mapping each input $\mathbf{x}$ to a distribution over the possible values of the latent representation $\mathbf{z}$ from which $\mathbf{x}$ could have been generated. ...
A neural network architecture which consists of two neural networks - a generator that creates data and a discriminator that evaluates it — trained adversarially to produce outputs that resemble real data. ...
Notes on using Gumbel-Softmax & Concrete Distribution in Categorical sampling ...