A study of adversarial attacks against malware detection

Abstract

The global volume of malware attacks has risen significantly over the last decade. A large majority of malware threats are aimed at the Windows operating system, leading to a greater demand for effective malware detection systems. Machine learning has been widely used in malware detection programmes to determine whether a file is malicious or benign. However, this approach is vulnerable to adversarial attacks, where the malware sample is incorrectly classified as a benign one. Moreover, in recent years, there has been an increase in the number of adversarial attacks on malware detection systems with attackers constantly finding new ways to evade detection. In this report, we provide an overview of the various types of adversarial attacks on malware detection models. Additionally, the paper will discuss existing research for such attacks on malware detection models. By evaluating the different adversarial attack methods and malware detection models and comparing their performances, we provide a justification for the differences in evasion rates. Finally, we conclude on the effectiveness of each method for malware detection, and their robustness to adversarial attacks.