Image Classification on CIFAR10 (test) (Robust Accuracy vs Perturbation)
46.95Robust Acc (8/255)PGD-10 (20)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PGD-10 (20)Backbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 46.95 | 33.3 | 22.29 | 11.48 | |
| Grad AlignBackbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 44.82 | 30.05 | 19.6 | 7.89 | |
| N-AAERBackbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 44.79 | 30.76 | 20.18 | 8.46 | |
| N-FGSMBackbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 44.43 | 30.32 | 19.06 | 6.78 | |
| MultiGradBackbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 44.39 | — | — | — | |
| ZeroGradBackbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 43.48 | — | — | — | |
| RS-AAERBackbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 43.22 | 26.75 | 17.03 | 5.37 | |
| RS-FGSMBackbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 43.17 | 0 | 0 | 0 | |
| PGD-2Backbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 42.97 | 28.63 | 18.52 | 3.77 | |
| FreeATBackbone=PreactResNet-18, Threat Model=L_inf, Evaluation Protocol=Auto Attack2024.04 | 40.23 | 28.04 | 0 | 0 |