A Comparison of State-of-the-Art Techniques for Generating Adversarial Malware Binaries

Abstract

We consider the problem of generating adversarial malware by a cyber-attacker where the attacker's task is to strategically modify certain bytes within existing binary malware files, so that the modified files are able to evade a malware detector such as machine learning-based malware classifier. We have evaluated three recent adversarial malware generation techniques using binary malware samples drawn from a single, publicly available malware data set and compared their performances for evading a machine-learning based malware classifier called Mal-Conv. Our results show that among the compared techniques, the most effective technique is the one that strategically modifies bytes in a binary's header. We conclude by discussing the lessons learned and future research directions on the topic of adversarial malware generation.

MalDBA: Detection for Query-Based Malware Black-Box Adversarial Attacks

“…We chose MalConv, a popular DL-based static malware detection model, which is used as the target model for many malware adversarial attacks [14,15,[30][31][32][33][34][35]. We reproduced the model using the Python programming language.…”

“…The function of this step is to train a mature malware detection model. For the DL-based static malware detector, we choose the MalConv model(as shown in Figure 3), which is not only the current popular malware detection model, but also the target model selected by many malware adversarial attacks [14,15,[30][31][32][33][34][35]. By training the MalConv model, a binary classifier that can distinguish benign samples from malicious samples can be obtained.…”