Rethinking Structural Anomaly Detection: From Decision Boundaries to Projection Operators
Researchers have proposed a new geometric approach to structural anomaly detection, moving beyond traditional probability density estimation or decision boundary methods. This novel technique learns a projection operator onto the manifold of normal samples, identifying anomalies based on the alteration caused by this projection. This method aims to improve performance by better aligning with the inductive bias of manifold-supported data and offers a unifying interpretation for reconstruction-based anomaly detection techniques. AI