This paper presents a unified, modular framework for graph matching and classification based on Graph Neural Networks. Based on GNN-learned node embeddings, we propose three complementary strategies for graph-level inference: (i) explicit node alignment via assignments of node embeddings, (ii) geometric abstractions of node embeddings using convex hull representations, and (iii) direct classification through nearest-neighbor aggregations of the embeddings. These approaches balance alignment accuracy and computational efficiency, enabling flexibility across applications. Experiments on molecular, biological, and social networks demonstrate competitive performance and highlight the framework’s interpretability, scalability, and runtime efficiency.
Dobler et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: