Adversarial vulnerability for any classifier

We prove the existence of adversarial perturbations that transfer across different classifiers.