Geometric Deep Learning?

I found a new website: http://geometricdeeplearning.com/
Good tutorials:

• Joan Bruna NIPS 2017 Geometric Deep Learning on Graphs and Manifolds
• PDF tutorial:https://arxiv.org/pdf/1611.08402.pdf
• Xavier Bresson: "Convolutional Neural Networks on Graphs"
• Slides: http://helper.ipam.ucla.edu/publications/dlt2018/dlt2018_14506.pdf

Basically in the tutorial Joan introduced the concept of manifold and graph theories. And then derived spectral graph convolution operations. What is missing is how can CNN be applied to general manifold instead of Euclidean space and graph?