Controlled graph neural networks with denoising diffusion for anomaly detection

Xuan Li, Chunjing Xiao, Ziliang Feng, Shikang Pang, Wenxin Tai, Fan Zhou

Abstract: Highlights•A diffusion model-based generator is proposed to manipulate node embeddings.•The neighborhood aggregation procedure in GNN is steered to generate augmented data.•A hypersphere-based semi-supervised model is presented to perform anomaly detection.