I'm very new to topology, however, I am working on a Convolutional Neural Network, and I stumbled upon topological neural networks.
There are very few papers on this topic, but from what I've read, topological neural networks modify the typical CNN framework by specifying the weights in the network to lie on a topological manifold such as the circle or Klein bottle.
I know this is a broad question, but how can I actually implement this? What would my filters (matrices) look like?
I'm just looking for a general description on how this would work. I read about Klein bottles although I'm not exactly sure how to create CNN layers out of this - i.e., to ensure my weights fall under the manifold.
Thank you very much.