Image Classification on MNIST (test) (Epsilon-based Robustness Evaluation)
62.6Accuracy (epsilon=3)Clip21-SGD2M
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Clip21-SGD2MModel=MLP, Hyperparameters=batch size 64, # epochs 150, n = 25, delta=10^-32025.02 | 62.6 | 75.9 | 83 | 87.7 | 89.2 | |
| Clip21-SGD2MModel=CNN, Hyperparameters=batch size 64, # epochs 150, n = 25, delta=10^-32025.02 | 61.2 | 76 | 80.9 | 87.6 | 89.6 | |
| Clip-SGDModel=MLP, Hyperparameters=batch size 64, # epochs 150, n = 25, delta=10^-32025.02 | 59.5 | 74.5 | 79.5 | 81.2 | 88.5 | |
| Clip-SGDModel=CNN, Hyperparameters=batch size 64, # epochs 150, n = 25, delta=10^-32025.02 | 58.9 | 78.7 | 82.8 | 83.9 | 91 | |
| Clip21-SGDModel=MLP, Hyperparameters=batch size 64, # epochs 150, n = 25, delta=10^-32025.02 | 49.2 | 68.1 | 79 | 77.9 | 86.7 | |
| Clip21-SGDModel=CNN, Hyperparameters=batch size 64, # epochs 150, n = 25, delta=10^-32025.02 | 46.1 | 67.9 | 76.4 | 79.3 | 86.7 | |
| α-NormECModel=CNN, Hyperparameters=batch size 64, # epochs 150, n = 25, delta=10^-32025.02 | 10.4 | 23 | 56.4 | 56.4 | 57.1 | |
| α-NormECModel=MLP, Hyperparameters=batch size 64, # epochs 150, n = 25, delta=10^-32025.02 | 9 | 28.7 | 42.2 | 53.4 | 64.1 |