Image Classification on SVHN (test) (Attributional Smoothness Metrics)
1.05Expected ||∇f(x)||^2Adversarial Training
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Adversarial TrainingBackbone=ResNet-182022.06 | 1.05 | 22.96 | 24.48 | 96.64 | |
| GradRegBackbone=ResNet-182022.06 | 1.63 | 68.22 | 39.55 | 96.57 | |
| LCNNs + GradRegBackbone=ResNet-182022.06 | 1.69 | 31.69 | 15.28 | 96.53 | |
| StandardBackbone=ResNet-182022.06 | 2.64 | 204.38 | 78.22 | 96.41 | |
| LCNNsBackbone=ResNet-182022.06 | 2.91 | 77.78 | 25.36 | 96.35 |