GAMBD: Generating adversarial malware against MalConv

Abstract

Deep learning technology has been widely used in malware detection in recent years. Although malware detection models based on deep learning are very effective in malware detection tasks, deep learning models are very vulnerable to adversarial attacks by adversarial examples. Adversarial malware can fool deep-learning malware detection models into classifying it as benign software. Due to the structural and discrete properties of Portable Executable (PE) files, it is very difficult and time-consuming to generate adversarial malware. Therefore, this paper proposes GAMBD, a gradient-based adversarial malware generative approach that can quickly generate powerful adversarial malware. Compared to existing methods, GAMBD has a higher success rate and takes less time to generate adversarial malware. Through experimental verification, GAMBD can achieve a success rate of 100% adversarial attack and the average time spent is only 2.8s, which is more than 30 times shorter than that of existing methods.