Variational Autoencoder: InfoVAE (or MMDVAE)

Implementation of the variational autoencoder InfoVAE and applied to MNIST digit dataset.

Structure

• The implementation consists of a PyTorch nn.Module which acts as the whole variational autoencoder.
• Then there are cells which contain code to generate new data (new synthetic images) and plot the distribution of the images and mean vectors in the learned latent space by the variational autoencoder.