Graph Data Augmentation

Graph data augmentation aims to improve the performance of graph neural networks (GNNs) by artificially increasing the size and diversity of training datasets, addressing challenges posed by data scarcity and high annotation costs. Current research focuses on developing novel augmentation techniques, including those based on generative models, diffusion processes, optimal transport methods (like Gromov-Wasserstein distances), and spectral graph properties, often integrated within contrastive learning frameworks. These advancements enhance GNN performance in various tasks, such as graph classification and anomaly detection, leading to more robust and generalizable models for applications across diverse domains.