defense arXiv Sep 12, 2025 · Sep 2025
Laith Nayal, Mahmoud Mousatat, Bader Rasheed · Innopolis University · LLC NUHA TECH
Defends ViT image classifiers against adversarial attacks using Lipschitz-constrained depth-dependent stochastic layer dropping
Input Manipulation Attack vision
Deep neural networks and Vision Transformers achieve state-of-the-art performance in computer vision but are highly vulnerable to adversarial perturbations. Standard defenses often incur high computational cost or lack formal guarantees. We propose a Lipschitz-guided stochastic depth (DropPath) method, where drop probabilities increase with depth to control the effective Lipschitz constant of the network. This approach regularizes deeper layers, improving robustness while preserving clean accuracy and reducing computation. Experiments on CIFAR-10 with ViT-Tiny show that our custom depth-dependent schedule maintains near-baseline clean accuracy, enhances robustness under FGSM, PGD-20, and AutoAttack, and significantly reduces FLOPs compared to baseline and linear DropPath schedules.
transformer Innopolis University · LLC NUHA TECH
ML Security Papers