Table 3 Accuracy against whitebox adversarial attacks on the MNIST dataset.
From: Evidence for the intrinsically nonlinear nature of receptive fields in vision
Attack methods | FGSM (\(\epsilon =0.1\)) | FGSM (\(\epsilon =0.2\)) | FGSM (\(\epsilon =0.3\)) | DeepFool | Carlini–Wagner (\(L_2\)) | Carlini–Wagner (\(L_\infty\)) |
|---|---|---|---|---|---|---|
CNN | 88.14% | 44.69% | 11.03% | 52.01% | 4.18% | 42.5% |
INRFnet | 93.14% | 62.23% | 33.42% | 65.27% | 7.24% | 58.06% |
