Fig. 6From: Deep neural rejection against adversarial examplesAdversarial examples computed on the CIFAR10 dataset adding a perturbation computed with ε=0.2. See the caption of Fig. 5 for further detailsBack to article page