StabilityAnalysisandGeneralizationBoundsof AdversarialTraining

In adversarial machine learning, deep neural networks can fit the adversarial examples on the training dataset but have poor generalizationability on the test set.