Deep graph learning for semi-supervised classification

• The global and local structure are jointly considered for deep graph learning networks. relationship of the structures mined by hierarchical progressive learning. different fusion dynamically encoded their interdependence. Graph (GL) can capture distribution (graph structure) data based on convolutional networks (GCN), quality directly influences GCN semi-supervised classification. Most existing methods combine computational layer related losses into exploring (measuring from all samples) or samples). emphasizes whole description inter-class data, while tends to neighborhood representation intra-class data. However, it is difficult simultaneously balance these process graphs classification because interdependence graphs. To simulate interdependence, (DGL) proposed find a better DGL not only learn previous metric computation updating, but also mine next weight reassignment. Furthermore, fuse encoding structures, deeply improve performance Experiments demonstrate that outperforms state-of-the-art three benchmark datasets (Citeseer, Cora, Pubmed) citation two (MNIST Cifar10) images.