Graphical-Based Learning Environments for Pattern Recognition

In this paper, we present a new neural network model, called graph neural network model, which is a generalization of two existing approaches, viz., the graph focused approach, and the node focused approach. The graph focused approach considers the mapping from a graph structure to a real vector, in which the mapping is independent of the particular node involved; while the node focused approach considers the mapping from a graph structure to a real vector, in which the mapping depends on the properties of the node involved. It is shown that the graph neural network model maintains some of the characteristics of the graph focused models and the node focused models respectively. A supervised learning algorithm is derived to estimate the parameters of the graph neural network model. Some experimental results are shown to validate the proposed learning algorithm, and demonstrate the generalization capability of the proposed model.