Anomaly Detection with Deep Compact Hypersphere
Yusuf Şalk, Bedirhan Uzun, Hakan Çevıkalp, Hasan Sarıbaş · 2022 30th Signal Processing and Communications Applications Conference (SIU) · 2022
The classifiers that return compact acceptance regions are crucial for the success in anomaly detection. This paper introduces a novel method that approximates the class acceptance regions with compact hypersphere models for anomaly detection. As opposed to the other deep hypersphere classifiers, we treat the hypersphere centers as learnable parameters and update them based on the changing deep feature representations. In addition, we propose a novel loss term that is more robust to label noises within outlier exposure dataset. The experimental results show that the proposed method achieves the state-of-the-art accuracies on the majority of the tested datasets in the context of anomaly detection.