A GAN Based Malware Adversaries Detection Model

Abstract

Deep Learning algorithms are effectively working for detection and classification in real-time systems. It surpasses human-level accuracy in image detection, disease classification, and many other fields. But recent studies show how deep learning detection systems are vulnerable to adversarial attacks. GANs are used to generate zero-day adversarial attacks by training the generator and discriminator network on a malware dataset. This study aims to provide a method to detect the malware adversaries generated by GAN. Firstly, we acquired a malware dataset from an online source. Secondly, a discriminator and generator network were selected to generate the adversarial data for testing purposes. In the end, we developed a novel deep neural network model and trained it using the augmented dataset. Our proposed model achieved an 84 % accuracy level in case of an adversary attack, and it forces the GAN network-based attack to create adversarial deformed samples. Our proposed model protects against deep learning-based adversarial attacks and helps in the detection of zero-day malware attacks.