Anomaly detection denotes the task of detecting anomalies in datasets usually in an unsupervised way. Anomaly refers to any instance that significantly differs from the other instances w.r.t. a certain property (e.g. an measurement outlier) or in a semantical sense like a cat-image in a pool of dog-images. Usually machine learning models for anomaly detection are trained on a clean set of examples and try to learn the underlying data distribution. When later confronted with an anomaly these models will identify the anomaly e.g. by returning a different metric value.