Abstract: Graph-level anomaly detection (GLAD) aims to identify graphs that significantly deviate from others in a graph dataset. Existing methods predominantly rely on standard Graph Neural Networks ...