Network of Graph Convolutional Networks \\ Trained on Random Walks

Anonymous

Nov 03, 2017 (modified: Nov 03, 2017) ICLR 2018 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: Semi-supervised learning on graph-structured data has recently progressed with the introduction of Graph Convolutional Networks (GCNs). At the same time, unsupervised learning of graph embeddings has benefited from random walks. In this paper, we marry the two, by feeding random walk statistics to multiple instances of GCNs, combining their output into a classification sub-network. Our overall architecture, Network of GCNs, is able to utilize information from near and distant neighbors. We evaluate on challenging node classification tasks, when the training data is very scarce, and we achieve state-of-the-art performance on Cora, Citeseer, and Pubmed. Further, we inspect how the classification sub-network circumvents adversarial input perturbations by shifting attention to distant nodes.
  • TL;DR: We make a network of Graph Convolution Networks, feeding each a different power of the adjacency matrix, combining all their representation into a classification sub-network, achieving state-of-the-art on semi-supervised node classification.
  • Keywords: Graph Convolution, Deep Learning, Network of Networks

Loading