Transfer Learning on Graph

Learning efficient graph techniques requires large, well-labeled datasets, which are rarely available across all domains. Transfer Learning on Graphs addresses this limitation by enabling knowledge learned from one graph or domain to be reused in another. This chapter presents a comprehensive view of Transfer Graph Learning (TGL), outlining its key features, benefits, and necessity in modern graph analytics. Inductive, transductive, and heterogeneous transfer learning paradigms are examined through representative algorithms, including GraphSAGE, GCN-DA, and Heterogeneous Feature Augmentation. This chapter further explains feature-based and structure-based transfer techniques, including Transfer Component Analysis, domain-adversarial learning, and meta-path-based methods. Case studies are also discussed based on the biological networks and anomaly detection.