Detecting anomalous network patterns

Spherical Defense is working on applying research from Cambridge University on representation learning and Natural Language Processing (NLP) to web application / API security. We learn the baseline in an unsupervised manner of normal JSON request headers and payloads, and can detect anomalies. We learn continuously as the application changes and reduce false positives. We can 1) detect anomalies in request using a tree-trace autoregressive model 2) Detect account takeover attacks using our tree-variational autoencoder model.