Spectral Graph Convolution Nerual Networks

Between August 2022 and Feb 2023, Dr Hwee Kuan Lee at the Bioinformatics Institute taught a course on graph neural networks to researchers at A*STAR. I helped teach a submodule within this course on the graph fourier transform and spectral graph convolution neural networks. You can find some of the slides here.

Broadly speaking, we covered the notion of signals on graphs, the relationship between the eigendecomposition of the normalized graph Laplacian and the structure of the graph, the notion of a Fourier transform on a graph (via the eigendecomposition), that of the convolution operation (via the FT), and finally the idea of graph convolutional neural networks. See the slides for details.