Malware-Induced Data Disturbance: A LeNet Perspective with FGSM and PGD

Malware, as a pervasive cyber threat, jeopardizes both digital infrastructure security and the broader societal functioning and safety. This paper investigates the resilience of the LeNet convolutional neural network (CNN) against malware-like adversarial perturbations on the Fashion MNIST dataset. Employing the Fast Gradient Sign Method (FGSM) and the Projected Gradient Descent (PGD) as representative attack vectors, simulate malware-induced data disturbances to evaluate their effect on LeNet’s performance. The experiments systematically assess the degradation in classification accuracy as a function of the intensity of adversarial noise. Results demonstrate a notable decline in model fidelity with increasing perturbation strength, accentuating LeNet’s vulnerability to such targeted data corruption. The research emphasizes the necessity for fortified defenses in machine learning infrastructures, particularly in applications where data integrity is critical. The findings of this research aim to catalyze advancements in the robustness of Convolutional Neural Networks (CNNs) against malware-exploited vulnerabilities in data-driven domains.