Graph Perturbation

Graph perturbation, the study of how graph-structured data changes affect machine learning models, primarily aims to improve the robustness and generalizability of graph neural networks (GNNs). Current research focuses on developing GNN architectures and algorithms that are resilient to node and edge perturbations, including methods leveraging pre-processing, continuous graph representations, and spectral regularization. This work is crucial for deploying GNNs in real-world applications where data is inherently noisy or subject to adversarial attacks, impacting fields like drug discovery, traffic forecasting, and social network analysis.