Deep Neural Networks achieve near-human performance on several benchmark datasets, yet they are not as robust as humans. Their success relies on the proximity of test samples to the distribution of training data, resulting ...