Image Classification on MNIST (Adversarial Robustness via PGD and Entropy)
10.18Clean AccuracyoffPure
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| offPureepsilon=2.02025.10 | 10.18 | 10.78 | 10.71 | 10.28 | 9.69 | |
| NNepsilon=2.02025.10 | 9.24 | 9.83 | 10.7 | 9.72 | 9.37 | |
| onPureepsilon=2.02025.10 | 3.46 | 3.79 | 4.1 | 3.68 | 3.38 | |
| OSepsilon=2.02025.10 | 2.9 | 3.89 | 5.62 | 3.91 | 3.09 | |
| ATepsilon=2.02025.10 | 2.82 | 7.68 | 14.9 | 5.54 | 3.24 | |
| NN50epsilon=2.02025.10 | 0.38 | 1.91 | 4.64 | 1.89 | 0.81 | |
| OS50epsilon=2.02025.10 | 0.37 | 1.7 | 3.56 | 1.65 | 0.81 | |
| MIXepsilon=2.02025.10 | 0.22 | 1.25 | 2.88 | 0.91 | 0.43 | |
| BLepsilon=2.02025.10 | 0.12 | 6.95 | 16.33 | 3.02 | 0.41 |