A Universal Growth Rate for Learning with Smooth Surrogate Losses