Summary of several graph neural network methods

Spatial Convolutional Network

The intuition of Spatial Convolution Network is similar to that of the famous CNN which dominates the literature of image classification and segmentation tasks. In short, the idea of convolution on an image is to sum the neighboring pixels around a center pixel, specified by a filter with parameterized size and learnable weight. Spatial Convolutional Network adopts the same idea by aggregate the features of neighboring nodes into the center node.


Left: Convolution on a regular graph such as an image. Right: Convolution on the arbitrary graph structure. Figure from “A Comprehensive Survey on Graph Neural Networks”

Graph Convolutional Networks (GCNs)

Convolutional neural networks (CNNs) have been vastly used for image classification and segmentation problems. Convolutional operation refers to applying a spatial filter to the input image and getting a feature map as a result. GCNs refer to applying a spatially moving filter over the nodes of the graph which contains embeddings or data relevant to each node to get a feature representation of each node. Stacking a number of convolutional layers like a regular CNN can also be done to incorporate information from larger neighborhoods.



Image source: https://www.experoinc.com/post/node-classification-by-graph-convolutional-network