Image Classification on CIFAR-10 (Certified Robustness Bounds)
78.6Clean Accuracyl2
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| l2Perturbation Budget (epsilon_inf, epsilon_2, epsilon_1)=(8/255, 0.5, 1.0)2024.10 | 78.6 | 0 | 56.5 | 75.8 | 0 | |
| CURE-RandomPerturbation Budget (epsilon_inf, epsilon_2, epsilon_1)=(8/255, 0.5, 1.0)2024.10 | 53 | 28.9 | 28 | 34.6 | 24 | |
| l_infinityPerturbation Budget (epsilon_inf, epsilon_2, epsilon_1)=(8/255, 0.5, 1.0)2024.10 | 51.8 | 36.3 | 6 | 3.8 | 3.5 | |
| CURE-MaxPerturbation Budget (epsilon_inf, epsilon_2, epsilon_1)=(8/255, 0.5, 1.0)2024.10 | 51.5 | 33.9 | 19.5 | 21.6 | 16.8 | |
| CURE-JointPerturbation Budget (epsilon_inf, epsilon_2, epsilon_1)=(8/255, 0.5, 1.0)2024.10 | 51.3 | 23.9 | 34 | 38.6 | 21.4 | |
| CURE-ScratchPerturbation Budget (epsilon_inf, epsilon_2, epsilon_1)=(8/255, 0.5, 1.0)2024.10 | 49.5 | 34.2 | 28.1 | 32 | 26.3 | |
| CURE-FinetunePerturbation Budget (epsilon_inf, epsilon_2, epsilon_1)=(8/255, 0.5, 1.0)2024.10 | 40.2 | 30.2 | 30.8 | 34.8 | 29.3 |