| Publisher | Springer |
| Authors | Huy Nguyen, Fabio Di Troia, Genya Ishigaki and Mark Stamp |
| Year | 2023 |
| Cost | Free |
| Themes | APTs, Attackers, Malware, Machine Learning, Malware Analysis, Malware Classification, Generative Adversarial Networks, Malware Detection |
Overview
For efficient removal of malware and accurate assessment of its threat level and potential damage, the classification of malware families is crucial. In this research paper, the authors propose a method that involves extracting features from malware executable files and representing them as images using various techniques.
They specifically focus on generative adversarial networks (GAN) for multiclass classification and compare the performance of their GAN model with other popular machine learning algorithms such as support vector machine (SVM), XGBoost, and restricted Boltzmann machines (RBM).
The study reveals that the discriminator of the AC-GAN model performs competitively compared to the other machine learning techniques. Additionally, the researchers investigate the effectiveness of the GAN generative model for adversarial attacks in image-based malware detection.
Despite the visually impressive images generated by the AC-GAN, the study finds that these generated images can be easily distinguished from real malware images using various learning techniques. This suggests that the GAN-generated images have limited value in adversarial attacks.
