Recently, I came across an article, Denoised Smoothing: A Provable Defense for Pretrained Classifiers, that suggested a method to deal with adversarial examples. The method consists of creating.