Generative Adversarial Network for Improving Deep Learning Based Malware Classification

The generative adversarial network (GAN) had been successfully applied in many domains in the past, the GAN network provides a new approach for solving computer vision, object detection and classification problems by learning, mimicking and generating any distribution of data. One of the difficulties in deep learning-based malware detection and classification tasks is lacking of training malware samples. With insufficient training data the classification performance of the deep model could be compromised significantly. To solve this issue, in this paper, we propose a method which uses the Deep Convolutional Generative Adversarial Network (DCGAN) to generate synthetic malware samples. Our experiment results show that by using the DCGAN generated adversarial synthetic malware samples, the classification accuracy of the classifier - a 18-layer deep residual network is significantly improved by approximately 6%.