Comment by esafak

Once you admit approximations the theoretical problem trades places with a more interesting one: what is the Pareto frontier of loss vs complexity?