Theoretical Foundations of Adversarially Robust Learning