Quantifying the Inductive Bias in Concept Learning