PhD Dissertation Proposal by QIAO Hezhe | Advancing Deep Learning for Graph Anomaly Detection: From Transductive Learning to Universal Generalist Models

Advancing Deep Learning for Graph Anomaly Detection:
From Transductive Learning to Universal Generalist Models

ABOUT THE TALK

Graph-structured data are pervasive in modern applications, including financial transaction networks, e-commerce user–item graphs, communication networks, cybersecurity logs, and social platforms. Graph anomaly detection (GAD) aims to identify irregular nodes that deviate from normal structural or feature patterns, which are widely used in fraud detection in finance, review spam detection, and abusive user detection. Despite significant advances in deep learning–based GAD, current methods predominantly face several challenges  (1) Overlooking the anomaly–discriminative property "one-class homophily" and the lack of a reliable anomaly scoring approach. (2) Ignoring the fact that  normal samples overwhelmingly dominate the data, making the labels for normal samples is much easier to obtain. (3) Limited generalization ability across different graphs, which restricts their effectiveness in real-world cross-graph applications.  To address these challenges, this dissertation proposed (1) a novel unsupervised anomaly scoring measure for GAD, local node affinity, and further propose Truncated Affinity Maximization (TAM) that learns tailored node representations. (2) a practical semi-supervised graph anomaly detection scenario, where part of the nodes in a graph are known to be normal and propose a novel Generative GAD approach (namely GGAD) for the semi-supervised scenario to better exploit the normal nodes. (3) AnomalyGFM, a GAD-oriented graph foundation model that supports zero-shot inference and few-shot prompt tuning for GAD in diverse graph datasets.

SPEAKER BIOGRAPHY

QIAO Hezhe is a fourth-year Ph.D. candidate in Computer Science at the School of Computing and Information Systems, under the supervision of Prof. Pang Guansong. Prior to that, he received his MS degree from the University of Chinese Academy Science, China. His research interests include multi-agent system safety,  graph anomaly detection,  LLM hallucination detection and mitigation, and graph representation learning.  He has published multiple CCF A papers as  (co)-first author, including NeurIPS x3, KDD x1, IJCAI x1, TKDE x1. His research has attracted 450+ citations and received multiple awards, e.g., SMU Dean List and President’s Award of the Chinese Academy of Sciences. He is also the main organizer and presenter of the IJCAI 2025 tutorial titled ‘Deep Learning for Graph Anomaly Detection’. and AAAI 2026 tutorial titled ‘Toward Foundation Models for Detecting Abnormal Activities on Graphs’. He also served on the program committees of multiple top conferences and journals.