On the use of high-order feature propagation in Graph Convolution Networks with Manifold Regularization

Graph Convolutional Networks (GCNs) have received a lot of attention in pattern recognition and machine learning. In this paper, we present revisited scheme for the new method called ”GCNs with Manifold Regularization” (GCNMR). While manifold regularization can add additional information, GCN-based semi-supervised classification process cannot consider full layer-wise structured information. Inspired by graph-based label propagation approaches, will integrate high-order feature into each GCN layer. High-order over graph fully exploit information provided latter at all GCN’s layers. It exploits clustering assumption, which is valid data but not well exploited GCNs. Our proposed would lead to more informative Using model, conduct several experiments on public image datasets containing objects, faces digits: Extended Yale, PF01, Caltech101 MNIST. We also three citation networks. The performs compared methods. With respect recent GCNMR approach, average improvements were 2.2%, 4.5%, 1.0% 10.6% MNIST, respectively.