LSGAN-AT: enhancing malware detector robustness against adversarial examples

Abstract Adversarial Malware Example (AME)-based adversarial training can effectively enhance the robustness of Machine Learning (ML)-based malware detectors against AME. AME quality is a key factor to enhancement. Generative Network (GAN) kind generation method, but existing GAN-based methods have issues inadequate optimization, mode collapse and instability. In this paper, we propose novel approach (denote as LSGAN-AT) ML-based detector Examples, which includes LSGAN module AT module. generate more effective smoother by utilizing brand-new network structures Least Square (LS) loss optimize boundary samples. makes using generated Robust Detector (RMD). Extensive experiment results validate better transferability in terms attacking 6 ML RMD resisting MalGAN black-box attack. The also verify performance recognition rate