Malware Detection Model Based on Deep Convolution Generation Adversarial Network

Abstract

At present, malware is one of the biggest threats to Internet security. In this paper, a new static malware analysis algorithm MSG is proposed based on DCGAN. The algorithm transforms the disassembled malware code into a gray image based on SimHash, and uses DCGAN to generate countermeasure samples for training to detect unknown malware variants. The experimental results show that the detection rate of our algorithm for malware can reach 96.67%, and the dodge rate of generated malicious samples can reach 0.92 under the detection of CNN discriminator.