Graph convolutional network (GCN) is one of the recent variants in graph neural networks (GNN) and efficiently utilizes the convolutional neural networks on graphs. Graph convolutional network (GCN) learns graph representation through multi-layers of neural networks followed by a non-linear activation function. The significant role of GCN is to produce a powerful feature representation of nodes in neural networks. GCN is a neural network architecture built to address the unnecessary complexity and redundant computation in deep learning models.
GCNs are categorized into Spectral graph convolutional networks and Spatial graph convolutional networks. Spatial-based methods operate on local graph neighborhoods to arbitrary graphs whereas, spectral-based methods use the spectral theory to attain the convolution operation on the topological graph. Spatial convolution is applied for the graphs like social networks, knowledge graphs, and molecular graphs. Spectral convolution deals with graph data of images videos in computer vision.
Applications fields of GCN include social network analysis, fraud detection, traffic prediction, computer vision, bio-informatics, applied chemistry and physics, materials science, natural language processing, and many more. Future developments of GCN are Deep graph convolutional networks, Graph convolutional networks for dynamic graphs, More powerful graph convolutional networks, and Multiple graph convolutional networks.