Graph Cross Networks with Vertex Infomax Pooling

An individual vertex is fully identified through its feature, which works as the vertex attribute. 's neighborhood, including both the internal connectivity information and contained vertex features Here we show more details about the graph datasets used in our experiments of both graph classification and vertex classification. Note that, three social network datasets, IMDB-B, IMDB-M and COLLAB do not provide specific vertex features, where the vertex dimension is denoted as 1 and the maximum vertex degrees are shown in addition. The detailed information of graph datasets used in the experiments of vertex classificationDataset Cora Citeseer Pubmed # V ertices 2708 3327 19717 # Edges 5429 4732 44338 # Classes 7 6 3 V ertex Dimension 1433 3703 500 # Train V ertices (full-sup.) The classification results are illustrated in Figure 1.