Randomized smoothing is a popular method to certify robustness of image classifiers to adversarial input perturbations. It is the only certification technique which scales directly to datasets of higher dimension such as ImageNet.