Graph Convolution Networks with manifold regularization for semi-supervised learning. (July 2020)

• Record Type: Journal Article
• Authors: Kejani, M. Tavassoli; Dornaika, F.; Talebi, H.
• Is Part Of: Neural networks. Volume 127(2020)
• Page Start: 160
• Page End: 167
• Publication Date: 2020-07
• Subjects: Graph-based semisupervised learning -- Graph Convolution Networks (GCN) -- Label prediction -- Manifold regularization -- Semisupervised image classification; Neural computers -- Periodicals; Neural networks (Computer science) -- Periodicals; Neural networks (Neurobiology) -- Periodicals; Nervous System -- Periodicals; Ordinateurs neuronaux -- Périodiques; Réseaux neuronaux (Informatique) -- Périodiques; Réseaux neuronaux (Neurobiologie) -- Périodiques; Neural computers; Neural networks (Computer science); Neural networks (Neurobiology); Periodicals; 006.32
• DOI: 10.1016/j.neunet.2020.04.016
• Languages: English
• ISSNs: 0893-6080

Abstract

Abstract: In recent times, Graph Convolution Networks (GCN) have been proposed as a powerful tool for graph-based semi-supervised learning. In this paper, we introduce a model that enhances label propagation of Graph Convolution Networks (GCN). More precisely, we propose GCNs with Manifold Regularization (GCNMR). The objective function of the proposed GCNMR is composed by a supervised term and an unsupervised term. The supervised term enforces the fitting term between the predicted labels and the known labels. The unsupervised term imposes the smoothness of the predicted labels of the whole data samples. By learning a Graph Convolution Network with the proposed objective function, we are able to derive a more powerful semi-supervised learning. The proposed model retains the advantages of the classic GCN, yet it can improve it with no increase in time complexity. Experiments on three public image datasets show that the proposed model is superior to the GCN and several competing existing graph-based semi-supervised learning methods. Highlights: We introduce a novel loss function for semi-supervised learning with GCNs. The loss function merges the cross-entropy term and the label smoothness term. The data graph is exploited in the two terms of the loss function. Label inference of the resulting GCN is enhanced. It is superior to the classic GCN and many competing semi-supervised methods.