Node Injection

Node injection attacks target the vulnerability of Graph Neural Networks (GNNs) by introducing malicious nodes into a graph, aiming to manipulate GNN predictions for tasks like node classification or graph classification. Current research focuses on developing effective attack strategies, often employing reinforcement learning or optimization-based approaches to generate realistic node features and connections, and exploring defenses against these attacks, including methods that leverage label uncertainty or orthonormal weight matrices. This research area is crucial because it highlights the security risks of deploying GNNs in real-world applications where data integrity is paramount, prompting the development of more robust and trustworthy GNN models.