Image Classification Robustness on CIFAR-10 (test)
88.26Accuracy (Natural)PGD-AT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PGD-ATdefense budget (epsilon_d)=8/255, Data augmentation=Cutout2022.01 | 88.26 | 49.23 | 39.77 | 39.25 | 40.38 | 37.61 | |
| PGD-ATdefense budget (epsilon_d)=8/2552022.01 | 88.07 | 47.93 | 37.61 | 36.96 | 38.58 | 35.44 | |
| PGD-ATdefense budget (epsilon_d)=8/255, Data augmentation=Random Noise2022.01 | 87.62 | 47.46 | 38.35 | 37.9 | 39.07 | 36.25 | |
| PGD-ATdefense budget (epsilon_d)=8/255, Data augmentation=AutoAugment2022.01 | 86.24 | 48.87 | 40.19 | 39.65 | 37.66 | 35.07 | |
| PGD-ATdefense budget (epsilon_d)=8/255, Data augmentation=Gaussian Smoothing2022.01 | 83.95 | 50.96 | 42.8 | 42.34 | 42.41 | 40.07 | |
| PGD-ATdefense budget (epsilon_d)=14/2552022.01 | 80 | 56.86 | 52.92 | 52.83 | 50.36 | 48.63 | |
| TRADESdefense budget (epsilon_d)=12/2552022.01 | 79.63 | 55.73 | 51.77 | 51.63 | 48.68 | 47.83 | |
| MARTdefense budget (epsilon_d)=14/2552022.01 | 77.29 | 57.1 | 53.82 | 53.71 | 49.03 | 47.67 |