Image Classification on CIFAR-10 original (test) (TPR-95 Robustness Metrics)
56.18TPR-95 Accuracy (Multi-target)PGD-AT + RR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PGD-AT + RRAT Framework=PGD-AT, Rejector=Rectified Rejection (RR), Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 56.18 | 57.57 | 54.08 | |
| PGD-AT + EBDAT Framework=PGD-AT, Rejector=EBD, Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 55.4 | 56.15 | 53.24 | |
| PGD-AT + SNetAT Framework=PGD-AT, Rejector=SNet, Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 55.02 | 55.79 | 51.37 | |
| TRADES + RRAT Framework=TRADES, Rejector=Rectified Rejection (RR), Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 54.83 | 57.93 | 51.48 | |
| TRADES + Con.AT Framework=TRADES, Rejector=Confidence, Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 53.69 | 56.89 | 50.88 | |
| CARL + MarginAT Framework=CARL, Rejector=Margin, Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 46.17 | 48.49 | 44.78 | |
| CCAT + RRAT Framework=CCAT, Rejector=Rectified Rejection (RR), Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 36.48 | 51.3 | 40.72 | |
| CCAT + Con.AT Framework=CCAT, Rejector=Confidence, Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 34.21 | 49.78 | 38.01 | |
| ATRO + MarginAT Framework=ATRO, Rejector=Margin, Backbone=ResNet-18, Threat Model=l_inf, eps=8/2552021.05 | 32.53 | 31.74 | 28.31 |