Deep learning methods have been recently extended to deal with graph-structured data. Instances of such geometric data are gene regulatory networks in genomics, brain structural network in neuroscience, sensor networks for high-energy physics, graph of products for recommender systems, molecules in quantum chemistry, knowledge graphs for NLP and computer vision, etc. In this tutorial, we will review some of the existing graph neural network architectures. Particularly, we will focus on ConvNets architectures based on spectral graph theory and spatial anisotropic diffusion processes. We will apply these networks to sub-graph matching, semi-supervised graph clustering, molecules property estimation and travelling salesman problem. These tasks will be implemented on Python notebooks during the tutorial.