Can we show that we need high dimensionality for learning?
I wonder if we can develop analogous theories to the complexity classes of computational automata.