Anomaly Detection Using a Variational Autoencoder, Part II

Let's start with a brief summary of the main ideas discussed in Part I. Industrial applications of anomaly detection are too many to list: fraud detection in banking, preventive maintenance in heavy industries, threat detection in cybersecurity, etc. In all these problems, defining outliers explicitly can be very challenging. Variational autoencoders (VAEs) automatically learn the general structure of the training data to isolate only its discriminative features, which are summarised in a compact latent vector. The latent vector constitutes an information bottleneck that forces the model to be very selective about what to encode. We train an encoder to produce the latent vector and a decoder to reconstruct the original data from the latent vector as faithfully as possible.