AnoOnly: Semi-supervised anomaly detection with the only loss on anomalies

Yixuan Zhou, Peiyu Yang, Yi Qu, Xing Xu, Zhe Sun, Andrzej Cichocki

Abstract: Highlights•A novel method named AnoOnly for the semi-supervised anomaly detection is proposed.•AnoOnly achieves balanced supervision between imbalanced normal and abnormal data.•AnoOnly replaces the loss supervision on normal data by a form of weak supervision.•This weak supervision is introduced by BN, which performs online cluster learning.