Ambiguous Model Learning Made Unambiguous with 1/f Priors