报告题目：

Beyond Accuracy: Graph Learning for Anomaly Detection

报告摘要：

Anomalies are not just outliers; they are often the most meaningful signals in data, pointing to fraudulent behavior in financial systems, irregular dynamics in brain activity, or unexpected disruptions in urban networks. Detecting such anomalies, however, is notoriously difficult. The patterns are often subtle, the data noisy, and the relationships deeply entangled within complex, evolving structures. Graphs provide a natural way to capture these interdependencies, and graph learning offers a powerful toolkit for advancing anomaly detection. In this talk, I will discuss how graph learning enables us to move beyond accuracy toward anomaly detection that is interpretable, robust, and human-centered. I will highlight key challenges facing this field and showcase recent solutions to these challenges. I will conclude by outlining a roadmap for the future of graph learning for anomaly detection, where the focus shifts from mere accuracy to a broader vision of scalability, interpretability, causality, and fairness. By integrating these dimensions, we can transform anomalies from being treated as noise into valuable opportunities for scientific discovery, system reliability, and actionable insights.