Improved MalGAN: Avoiding Malware Detector by Leaning Cleanware Features

Abstract

In recent years, researches on malware detection using machine learning have been attracting wide attention. At the same time, how to avoid these detections is also regarded as an emerging topic. In this paper, we focus on the avoidance of malware detection based on Generative Adversarial Network (GAN). Previous GAN-based researches use the same feature quantities for learning malware detection. Moreover, existing learning algorithms use multiple malware, which affects the performance of avoidance and is not realistic on attackers. To settle this issue, we apply differentiated learning methods with the different feature quantities and only one malware. Experimental results show that our method can achieve better performance than existing ones.