Investigating Power laws in Deep Representation Learning