Image Classification on CIFAR100 (test) (L2 Perturbation Robustness)
77.44AccuracySoftplus + Wt. Decay
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Softplus + Wt. DecayBackbone=ResNet-18, Attack=l2 PGD, Evaluation=averaged across two runs2022.06 | 77.44 | 60.86 | 38.04 | 23.85 | 16.2 | |
| StandardBackbone=ResNet-18, Attack=l2 PGD, Evaluation=averaged across two runs2022.06 | 77.42 | 59.97 | 37.55 | 23.41 | 16.11 | |
| LCNNS + GradRegBackbone=ResNet-18, Attack=l2 PGD, Evaluation=averaged across two runs2022.06 | 77.29 | 72.68 | 63.36 | 52.96 | 42.7 | |
| GradRegBackbone=ResNet-18, Attack=l2 PGD, Evaluation=averaged across two runs2022.06 | 77.2 | 71.9 | 61.06 | 49.19 | 38.09 | |
| LCNNBackbone=ResNet-18, Attack=l2 PGD, Evaluation=averaged across two runs2022.06 | 77.16 | 61.17 | 39.72 | 25.6 | 17.66 | |
| Adversarial TrainingBackbone=ResNet-18, Attack=l2 PGD, Evaluation=averaged across two runs2022.06 | 76.96 | 72.76 | 64.7 | 54.8 | 44.98 | |
| CUREBackbone=ResNet-18, Attack=l2 PGD, Evaluation=averaged across two runs2022.06 | 76.48 | 71.39 | 61.28 | 49.6 | 39.04 |