Utilizing Generative Adversarial Networks to develop a robust Defensive System against Adversarial Examples

Abstract: Using the special ability of Generative Adversarial Networks (GANs) to create fresh adversarial instances for model retraining, we offer a novel defense strategy against adversarial examples in this study. In order to achieve this, we create an automated pipeline that combines a convolutional neural network that has already been trained with an external GAN called the Pix2Pix conditional GAN. This pipeline allows us to identify the transformations between adversarial examples and clean data as well as create new adversarial examples on the fly. In an iterative pipeline, these adversarial samples are used to strengthen the model, attack, and defense. Our simulation findings show that the suggested strategy works well.

Keywords: adversarial machine learning; botnet detection; generative adversarial networks; machine learning