PGD (Projected Gradient Descent)

anchor

Iterative adversarial attack that repeatedly steps along the loss gradient and projects back into the allowed perturbation ball.

Note

anchor found by search and checked against this entry's own description: "Towards Deep Learning Models Resistant to Adversarial Attacks", which motivates projected gradient descent as the universal first-order adversary this entry describes

Findings

• IC-028SPADE, an abstaining classifier built on top of ResNet, ViT, and VGG models, detects out-of-distribution and adversarial samples with provable guarantees. [eval]
• IC-077VGG-16's loss landscape barrier height distinguishes adversarial from real inputs, enabling a detection method that outperforms baselines on DeepFool and C&W attacks [eval]
• IC-1195Model substitution adversarial attack reduces TreeRing AUROC to 0.14 at ε=2/255 and StegaStamp AUROC to 0.492 at ε=12/255 [primary]
• IC-542The PyTorch pretrained ResNet50 on ImageNet is vulnerable to (1, y)-ACE calibration attacks that increase ECE from 3.70% to 47.23% while preserving accuracy [supporting]