The sort of mathematics we can do with NNs is to take limits of layer sizes such that the start to behave like GPs. Then we understand how the NN is working in that limit.

→ View original post on X — @lawrennd,