PhD Dissertation Proposal by LIU Zhonghang | Statistical-based Solutions to Unsupervised Visual Anomaly Detection

Unsupervised visual anomaly detection is crucial in various real-world application domains such as defect detection and it is essential for enhancing the robustness of machine learning algorithms. Although there is a plethora of image anomaly detectors available today, most of them focus primarily on feature representation learning, while underestimating the significance of data pre-processing (normalization) and intrinsic distribution mining (multi-normality learning), which can significantly impact algorithm performance. In this work, we address these limitations and propose an effective framework. Besides, we introduce a novel perspective: threshold learning for wider applications.

LIU Zhonghang is a Ph.D. candidate in Computer Science at the SMU School of Computing and Information Systems, supervised by Assistant Professor Daniel LIN. His research interests focus on adopting statistical machine learning methods on some computer vision tasks, especially for unsupervised visual anomaly detection.