Table 2 Adversarial trained model performance of perturbed and unperturbed data
From: On the role of deep learning model complexity in adversarial robustness for medical images
Attack | Resnet-8 | Resnet-20 | Resnet-32 | Resnet-50 |
|---|---|---|---|---|
(a) Chest X-Ray Accuracy (%),\(\epsilon\) = 10 | ||||
No Attack | 94.01 | 88.27 | 84.92 | 89.90 |
PGD | 82.14 | 74.05 | 72.96 | 61.11 |
(b) Dermoscopy Accuracy (%),\(\epsilon\) = 1 | ||||
No Attack | 88.14 | 89.56 | 88.02 | 89.64 |
PGD | 82.37 | 74.49 | 80.01 | 74.92 |
(c) OCT Accuracy(%),\(\epsilon\) = 4 | ||||
No Attack | 78.12 | 71.01 | 62.77 | 88.67 |
PGD | 47.28 | 62.56 | 57.06 | 67.95 |