> By minimizing a loss functional with respect to a bunch of numbers that amount to entries in matrices (or tensors, whatever) using an approximate hill climbing approach.

Are the rules of chess all there is to it? Is there really no more to be said?