Geometric deep learning is an umbrella term for emerging techniques attempting to generalize (structured) deep neural mod - els to non-Euclidean domains, such as graphs and manifolds.

geometricdeeplearning.com

See this review article

Datasets

https://medium.com/ai%C2%B3-theory-practice-business/abc-free-datasets-for-geometric-deep-learning-5e2995768b37 https://deep-geometry.github.io/abc-dataset/

Supervised learning (learning functions) on Non-Euclidean domains

We can further break down such problems into two subclasses: problems where the domain is fixed and those where multiple domains are given.

Graph convolutional networks (see Convolutional neural networks)

In computer graphics and vision applications, finding similari- ty and correspondence between shapes are examples of the second subclass of problems

Masci et al. [47] showed the first CNN model on meshed surfaces, resorting to a spatial definition of the convo - lution operation based on local intrinsic patches.

Manifold learning