Learning with Monotone Adversarial Corruptions