💠 Edge#201: Understanding Graph Convolutional Neural Networks
In this issue:
we explain Graph Convolutional Neural Networks;
we overview the original GCN Paper;
we explore PyTorch Geometric, one of the most complete GNN frameworks available today.
Enjoy the learning!
💡 ML Concept of the Day: Understanding Graph Convolutional Neural Networks
Continuing our series about graph neural networks (GNNs), today we would like to discuss one of its most popular variations. Graph convolutional networks (GCNs) were presented in 2017 in a paper from AI researchers from the University of Amsterdam and quickly became one of the most popular GNN methods in research.
As its name indicates, GCNs are a variation of convolutional neural networks (CNNs) but applied to graph datasets. The idea sounds simpler than it really is in practice. Traditional convolution operations work well in structures such as pixel vectors in which the positions of pixels are absolute. This contrast with a graph mode in which the neighborhood structure changes from node to node. To address this challenge,