Graph Convolutional Neural Network (GCNN) | Explained with a simple numerical example

“Classifying research papers using Graph Neural Network”: Convolution operation in images helps us identify features from the image by considering not just the value of individual pixels, but also the values of neighboring pixels. It is also possible to extract meaning from the features using a convolution operation in Graph Neural Networks (GNN). Imagine a Graph Neural Network (GNN) that represents research papers. Each paper cites a few other papers in the network. The papers are represented by nodes and citations by edges. The individual papers may be based on a few different topics: say Machine Learning, NLP, GenAI, Computer Vision etc. Often when we are referring to a paper, we are very interested to find similar papers. Is it possible to group the papers into a few different categories using the framework of GNN? This is totally possible. To classify the papers (nodes of the GNN), we consider the features of the papers (nodes) as well as those of the neighboring nodes (features of cited papers). We can accomplish feature aggregation and node classification using a concept called Graph Convolutional Network (GCN). This lecture teaches you how to use GCN to perform a simple node classification operation, using a numerical example. The example is very easy to understand. If you are a total beginner or have no idea about GNNs I am quite certain that this lecture will still help you. I build the GNN from scratch and hand-calculate the message passing, feature aggregation, transformation and node classification. Very simple. Please check out the lecture here. I have contributed this lecture is part of the Graph Neural Network lecture organized by Aiswarya.