Pytorch implementation of Variational Autoencoders - a popular deep generative probabilistic graphical model.
To train: python train.py
To sample from the saved model: python sample.py
Variational Autoencoders (VAEs) are true generative models which differ from the regular autoencoders in a sense that a VAE generates a probability distribution over the hidden variable z given the input data x, i.e., P(z/X), rather than just learning the deterministic mapping from input data to hidden representation as is the case with regular autoencoders. The goal is to learn the probability distribution over the observed variable X, and this is facilitated by learning to generate the distribution P(X/z).