Animation: Variational Autoencoder

A variational autoencoder is a type of neural network that learns to compress (encode) data, in such a way that one can later randomly sample data.

Here we show training of a variational autoencoder on a data set of wave packets that have been placed with a random location and amplitude. Upper left: test set, with VAE decoder results shown in orange; upper right: latent space distribution, with color coding for actual location (left) or amplitude (right) of any data point; lower right: latent space distributions for the few samples from the test set (in each case, the Gaussian spread in latent space is also indicated); lower left: loss evolution.