Spectral methods are instrumental for the representation of large graphs in vector spaces of low dimension. In this talk, we will review these methods, their physical interpretation, and highlight the impact of some pre- and post-processing steps (regularization, scaling, normalization). We will also show how to apply these techniques using scikit-network, a Python package for the analysis of large graphs.
Keywords: Graph embedding, Laplacian matrix, spectral decomposition, random projection.