Learning-based pattern classifiers, including deep networks, have shown
impressive performance in several application domains, ranging from computer
vision to cybersecurity. However, it has also been shown that adversarial input
perturbations carefully crafted either at training or at