AMGmal: Adaptive mask-guided adversarial attack against malware detection with minimal perturbation

Maleficent Neural Networks, the Embedding of Malware in Neural Networks: A Survey

“…Addressing challenges posed by traditional methods in adapting to massive data [12] AMGmal for resilience [7] Resilience in Adversarial Machine Learning Advocating for resilience in malware detection models against adversarial attacks. [2] Challenges in anti-malware technologies Emphasis on adaptive strategies due to developments in encryption and obfuscation [8] Resource conservation and accuracy maintenance Focus on feature selection and hybrid deep learning frameworks [1] DQEAF framework for evading traditional detection engines…”

“…Use of reinforcement learning for evading detection engines [9] Lightweight CNN for efficient feature extraction Utilisation of disassembled instructions for identifying polymorphic and zero-day malware [10] Multilayered ANN for comprehensive understanding Incorporation of data from various sources for a comprehensive understanding [11] Deep learning with feature extraction Use of grayscale images and OpCode 3-grams for automated and adaptable malware characterisation [12] AMGmal for adversarial attack mitigation additions to the literature for this research, as understanding the techniques that can be employed is crucial for contributing to further studies aimed at finding ways to detect or mitigate these "evil" models.…”

“…Zhan et al [12] address the vulnerability of deep neural networks to adversarial examples in malware detection, presenting AMGmal, that is, adaptive mask-guided adversarial attack against malware detection with minimal perturbation, an innovative strategy that uses saliency detection and mask guidance to prioritise critical bytes in slack areas. Experimental results demonstrate AMGmal's effectiveness in evading detection and reducing perturbations.…”

“…Zhan et al [12] introduce AMGmal, a mitigation strategy focusing on adversarial attacks on DNN-based detectors. The strategy involves fooling visualisation-based detectors, reserving functionality, and minimising perturbation, contributing to enhanced robustness against adversarial attacks.…”

“…As part of our research, we have found that adversarial attacks require innovative frameworks and countermeasures. Using AMGmal by Zhan et al [12], a mitigation strategy aimed at adversarial attacks against DNN-based detectors, is an example of these forward-looking approaches required in today's cyber environment.…”