Multi-Domain Generalized Graph Meta Learning

Graph meta learning aims to learn historical knowledge from training graph neural networks (GNNs) models and adapt it downstream tasks in a target graph, which has drawn increasing attention due its ability of transfer fast adaptation. While existing approaches assume the are same domain but lack solution for multi-domain In this paper, we address generalized problem, is challenging non-Euclidean data, inequivalent feature spaces, heterogeneous distributions. To end, propose novel called MD-Gram generalization. It introduces an empirical generalization method that uses vectors form unified expression data. Then proposes graphs transformation approach transform multiple source-domain with spaces into common domain, where conducted knowledge. further adopts domain-specific GNN enhancement customized model achieve adaptation unseen domain. Extensive experiments based on four real-world datasets show proposed significantly outperforms state-of-the-art tasks.