Unsupervised surrogate anomaly detection

Total: 1

In this paper, we study an unsupervised anomaly detection algorithm that learns neural network representations, that is, anomalies deviate from regular patterns of normal data. Inspired by similar concepts in engineering, we call our methodology surrogate anomaly detection. We formalize the concept of surrogate anomaly detection as a set of axioms required for an optimal surrogate model and propose a new algorithm, called Deep Ensemble ANomaly Detection (DEAN), designed to meet these criteria. We evaluated DEAN on 121 benchmark datasets, demonstrating competitive performance against 19 existing methods as well as the scalability and reliability of our method.