Defense-GAN: Protecting Classifiers Against Adversarial Attacks Using Generative Models

In recent years, deep neural network approaches have been widely adopted for learning tasks, including classification. However, they were shown to vulnerable to adversarial perturbations: carefully crafted small can cause misclassification of legitimate images. We propose-GAN, a new framework leveraging the expressive capability of generative to defend deep neural networks against such attacks. Defense-GAN is to model the distribution of unperturbed images. At inference time, it a close output to a given image which does not contain the adversarial. This output is then fed to the classifier. Our proposed method can be with any classification model and does not modify the classifier structure training procedure. It can also be used as a defense against any attack as does not assume knowledge of the process for generating the adversarial. We empirically show that Defense-GAN is consistently effective different attack methods and improves on existing defense strategies. code has been made publicly available at: //github. com/kabkabm/defensegan