Anomaly Detection with Variational AutoEncoder in TensorFlow for Deep Learning course @TU Eindhoven
Variational Autoencoders (VAEs) provide a mathematically grounded framework for the unsupervised learning of latent representations. It is possible to perform unsupervised anomaly detection by training a VAE on the training data, such that it learns to represent "normal" data well and then compute the ELBO values for the test data, where ideally "normal" examples should obtain higher likelihood values than anomalous examples.
Assignment worked out with the collaboration of a team member.