Image Classification on CIFAR-10 (test) (Certified Robustness L1/L2/Linf)
82.1Clean Accuracyl2
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| l2epsilon_inf=2/255, epsilon_2=0.25, epsilon_1=0.52024.10 | 82.1 | 5.8 | 71.3 | 81.7 | 5.8 | |
| CURE-Jointepsilon_inf=2/255, epsilon_2=0.25, epsilon_1=0.52024.10 | 79.4 | 56.2 | 68.1 | 77.1 | 56.2 | |
| l∞epsilon_inf=2/255, epsilon_2=0.25, epsilon_1=0.52024.10 | 79.2 | 60.3 | 67.3 | 75.9 | 60.3 | |
| CURE-Randomepsilon_inf=2/255, epsilon_2=0.25, epsilon_1=0.52024.10 | 78.4 | 59 | 68.5 | 76.9 | 58.9 | |
| CURE-Finetuneepsilon_inf=2/255, epsilon_2=0.25, epsilon_1=0.52024.10 | 78 | 59.7 | 68.2 | 75.9 | 59.7 | |
| CURE-Maxepsilon_inf=2/255, epsilon_2=0.25, epsilon_1=0.52024.10 | 77.6 | 60 | 69.3 | 75.2 | 60 | |
| CURE-Scratchepsilon_inf=2/255, epsilon_2=0.25, epsilon_1=0.52024.10 | 76 | 61.2 | 67.7 | 74.6 | 61.2 |