Towards Semi-supervised Universal Graph Classification

Graph neural networks have pushed state-of-the-arts in graph classifications recently. Typically, these methods are studied within the context of supervised end-to-end training, which necessities copious task-specific labels. However, real-world circumstances, labeled data could be limited, and there a massive corpus unlabeled data, even from unknown classes as complementary. Towards this end, we study problem semi-supervised universal classification, not only identifies samples do belong to known classes, but also classifies remaining into their respective classes. This is challenging due severe lack labels potential class shifts. In paper, propose novel network framework named UGNN, makes best subgraph perspective. To tackle shifts, estimate certainty graphs using multiple subgraphs, facilities discovery categories. Moreover, construct semantic prototypes embedding space for both categories utilize posterior prototype assignments inferred Sinkhorn-Knopp algorithm learn abundant across different views. Extensive experiments on six datasets verify effectiveness UGNN settings.