A General Framework for Adaptive Anomaly Detection with Evolving Connectionist Systems
Yihua Liao, V. Rao Vemuri, Alejandro Pasos Ruiz · 2004
Anomaly detection techniques hold great potential for combating novel intrusions, malicious insiders and fast spreading worms. A widely acknowledged challenge in anomaly detection is how to accurately model a subject's normal behavior in the presence of concept drift. This paper presents a new adaptive anomaly detection framework that aims to adapt to normal behavior changes while still recognizing anomalies. Evolving connectionist systems are employed to learn a subject's behavior in an online, adaptive fashion without a priori knowledge of the underlying data distributions.