Anomaly Detection with Complex Data Structures

Abstract

Identifying anomalies with complex patterns is different from the conventional anomaly detection problem. Firstly, for cross-modal anomaly detection problems, a large portion of data instances within a multi-modal context is often not anomalous when they are viewed separately in each modality, but they present abnormal patterns or behaviors when multiple sources of information are jointly considered and analyzed. Secondly, for the attribution network anomaly detection problem, the definition of anomaly becomes more complicated and obscure. Apart from anomalous nodes whose nodal attributes are rather different from the majority reference nodes from a global perspective, nodes with nodal attributes deviate remarkably from their communities are also considered to be anomalies. Thirdly, given a specific task with the different data structures, the process of building a suitable and high-quality deep learning-based outlier detection system still highly relies on human expertise and laboring trials. It is also necessary to automatically search the suitable outlier detection models for different tasks. In this dissertation, we made a series of contributions to enable advanced anomaly detection techniques for complex data structures and discussing how to automatically design anomaly detection frameworks for various data structures.