This repository provides a Tensorflow 2 implementation of some GNN layers using tf.keras.layers.Layer
subclassing, as well a notebook to evaluate their expressive power through an image filtering task.
The general framework for spatial graph convolution (message passing) is :
A GNN layer can be defined by different C matrices. For now, this project contains :
Muhammet Balcilar, Guillaume Renton, Pierre Héroux, Benoit Gaüzère, Sébastien Adam, Paul Honeine. Analyzing the Expressive Power of Graph Neural Networks in a Spectral Perspective. ICLR 2021.