On The Interplay Between Scale And Inductive Bias In Deep Learning