# Graph Neural Network **Entity class:** Machine-learning model family **Aliases:** GNN **Maturity:** Developed ## Definition A **graph neural network** learns representations by aggregating information from nodes, edges, and neighborhoods. In threat analysis it can support classification, anomaly detection, and link prediction, but its outputs inherit missingness, labeling bias, and adversarial changes in the source graph. ## Relationships - **used for:** [[wiki/Link Prediction|Link Prediction]] and [[wiki/Graph Analytics|Graph Analytics]]. - **families include:** [[wiki/Graph Convolutional Network|Graph Convolutional Network]] and [[wiki/GraphSAGE|GraphSAGE]]. - **adversarial case:** [[wiki/SAVAGE|SAVAGE]]. ## Sources / Provenance - [[research/A Network-Epidemiological Analysis of Radicalization and Real-Time Counterterrorism Observability|Network-Epidemiological Analysis]]