Wow, that paper is 45 MB. I read the article and I now looking at the paper, trying to decide whether to spend a day really absorbing it. I assume that you are recommending it. I really like the typography, use of margin text and math notation, etc.
Here is also a talk about this topic, that Petar Veličković gave virtually at the University Erlangen a few weeks ago (and why it is called Erlangen Programme)
This has nothing in common with Erlang Programming Language, it's about "method of characterizing geometries based on group theory and projective geometry." https://en.wikipedia.org/wiki/Erlangen_program
5 comments
[ 128 ms ] story [ 34.3 ms ] threadGeometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges
https://arxiv.org/abs/2104.13478
Stanford CS224W: Machine Learning with Graphs
https://www.youtube.com/playlist?list=PLoROMvodv4rPLKxIpqhjh...
https://www.youtube.com/watch?v=9cxhvQK9ALQ