Universality of shallow and deep neural networks on non-Euclidean spaces

Vugar Ismailov

We study shallow and deep neural networks whose inputs range over a general topological space. The model is built from a prescribed family of continuous feature maps and reduces to…