Machine learning just feels like a bunch of people trying random shit until something kind of works. Granted, that's probably the very definition of this kind o

Machine learning just feels like a bunch of people trying random shit until something kind of works. Granted, that's probably the very definition of this kind of exercise in approximation, but I'd like there to be a bit more formalism in the creation, and not just in the validation, of these networks. It's like computer science is hardly even a science and more of an art when you come to ML - that's not a good thing!