Classification on FashionMNIST l1, epsilon=10 (test)
92.56Clean AccuracyDense Orthogonal Classifier
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Dense Orthogonal Classifieralpha=12021.05 | 92.56 | 89.81 | 87.9 | |
| FGSM-AT2021.05 | 91.94 | 20.72 | 12.75 | |
| Standard Training2021.05 | 91.82 | 24.5 | 8.27 | |
| TRADES1/lambda=12021.05 | 86.06 | 68.88 | 58.4 | |
| ALP2021.05 | 83.08 | 65.74 | 53.65 | |
| Dense Orthogonal Classifieralpha=0.32021.05 | 78.8 | 70.1 | 51.81 | |
| TRADES1/lambda=62021.05 | 78.05 | 63.98 | 50.34 | |
| PGD-AT2021.05 | 77.76 | 67.2 | 57.99 |