Image Classification on CIFAR-10 (test) (Robustness)
91.9RobustnessPCL
Evaluation results
| Method | Links | |
|---|---|---|
| PCLAdversarial Training=PGD, Attack=No Attack2019.04 | 91.9 | |
| PCLAdversarial Training=FGSM, Attack=No Attack2019.04 | 91.3 | |
| PCLAdversarial Training=PGD, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 91.3 | |
| PCLAdversarial Training=FGSM, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 91 | |
| BaselineAdversarial Training=No, Attack=No Attack2019.04 | 90.8 | |
| Pang et al.Adversarial Training=Yes, Attack=No Attack2019.04 | 90.6 | |
| PCLAdversarial Training=No, Attack=No Attack2019.04 | 90.5 | |
| Madry et al.Adversarial Training=Yes, Attack=No Attack2019.04 | 87.3 | |
| Ross et al.Adversarial Training=Yes, Attack=No Attack2019.04 | 86.2 | |
| AdvTrainAdversarial Training=Yes, Attack=No Attack2019.04 | 84.5 | |
| Madry et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 84.5 | |
| PCLAdversarial Training=No, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 84.3 | |
| Yu et al.Adversarial Training=Yes, Attack=No Attack2019.04 | 83.1 | |
| Yu et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 82.5 | |
| PCLAdversarial Training=FGSM, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 80.8 | |
| Pang et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 80.6 | |
| PCLAdversarial Training=PGD, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 78.5 | |
| PCLAdversarial Training=PGD, Attack=PGD, Perturbation Budget (epsilon)=0.012019.04 | 75.7 | |
| PCLAdversarial Training=PGD, Attack=MIM, Perturbation Budget (epsilon)=0.012019.04 | 74.9 | |
| PCLAdversarial Training=PGD, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 74.5 | |
| PCLAdversarial Training=PGD, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 73.7 | |
| PCLAdversarial Training=FGSM, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 72.9 | |
| PCLAdversarial Training=No, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 72.5 | |
| Ross et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 72.2 | |
| Madry et al.Adversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 71.6 | |
| PCLAdversarial Training=FGSM, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 70.5 | |
| PCLAdversarial Training=PGD, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 69.9 | |
| PCLAdversarial Training=FGSM, Attack=MIM, Perturbation Budget (epsilon)=0.012019.04 | 68.8 | |
| PCLAdversarial Training=FGSM, Attack=PGD, Perturbation Budget (epsilon)=0.012019.04 | 68.3 | |
| PCLAdversarial Training=FGSM, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 67.9 | |
| AdvTrainAdversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 67.7 | |
| Madry et al.Adversarial Training=Yes, Attack=PGD, Perturbation Budget (epsilon)=0.012019.04 | 67.7 | |
| Madry et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 65.7 | |
| Madry et al.Adversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 64.3 | |
| PCLAdversarial Training=No, Attack=MIM, Perturbation Budget (epsilon)=0.012019.04 | 64.3 | |
| PCLAdversarial Training=No, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 63.5 | |
| PCLAdversarial Training=No, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 62.9 | |
| Yu et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 62.9 | |
| Yu et al.Adversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 62.7 | |
| Pang et al.Adversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 61.7 | |
| Madry et al.Adversarial Training=Yes, Attack=MIM, Perturbation Budget (epsilon)=0.012019.04 | 61.5 | |
| BaselineAdversarial Training=No, Attack=C&W, Perturbation Budget (c)=0.0012019.04 | 61.3 | |
| PCLAdversarial Training=PGD, Attack=C&W, Perturbation Budget (c)=0.12019.04 | 60.5 | |
| PCLAdversarial Training=No, Attack=PGD, Perturbation Budget (epsilon)=0.012019.04 | 60.1 | |
| PCLAdversarial Training=PGD, Attack=MIM, Perturbation Budget (epsilon)=0.022019.04 | 60 | |
| PCLAdversarial Training=PGD, Attack=PGD, Perturbation Budget (epsilon)=0.022019.04 | 58.5 | |
| PCLAdversarial Training=PGD, Attack=BIM, Perturbation Budget (epsilon)=0.022019.04 | 57.3 | |
| PCLAdversarial Training=No, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 56.3 | |
| PCLAdversarial Training=FGSM, Attack=C&W, Perturbation Budget (c)=0.12019.04 | 55.7 | |
| Pang et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 54.9 | |
| PCLAdversarial Training=FGSM, Attack=MIM, Perturbation Budget (epsilon)=0.022019.04 | 53.8 | |
| Pang et al.Adversarial Training=Yes, Attack=MIM, Perturbation Budget (epsilon)=0.012019.04 | 52.1 | |
| PCLAdversarial Training=FGSM, Attack=BIM, Perturbation Budget (epsilon)=0.022019.04 | 51.2 | |
| PCLAdversarial Training=FGSM, Attack=PGD, Perturbation Budget (epsilon)=0.022019.04 | 50.6 | |
| Madry et al.Adversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.022019.04 | 49.3 | |
| Yu et al.Adversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 48.5 | |
| Madry et al.Adversarial Training=Yes, Attack=PGD, Perturbation Budget (epsilon)=0.022019.04 | 48.5 | |
| Pang et al.Adversarial Training=Yes, Attack=PGD, Perturbation Budget (epsilon)=0.012019.04 | 48.4 | |
| Madry et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.12019.04 | 47.9 | |
| Ross et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 47.8 | |
| Madry et al.Adversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 47.4 | |
| Madry et al.Adversarial Training=Yes, Attack=MIM, Perturbation Budget (epsilon)=0.022019.04 | 46.7 | |
| Pang et al.Adversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 46.6 | |
| Pang et al.Adversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 46.2 | |
| AdvTrainAdversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 44.3 | |
| PCLAdversarial Training=No, Attack=MIM, Perturbation Budget (epsilon)=0.022019.04 | 42.3 | |
| PCLAdversarial Training=No, Attack=C&W, Perturbation Budget (c)=0.12019.04 | 41.1 | |
| AdvTrainAdversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 40.9 | |
| Yu et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.12019.04 | 40.7 | |
| PCLAdversarial Training=No, Attack=BIM, Perturbation Budget (epsilon)=0.022019.04 | 40.1 | |
| Ross et al.Adversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 39.5 | |
| Yu et al.Adversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.022019.04 | 39.3 | |
| PCLAdversarial Training=No, Attack=PGD, Perturbation Budget (epsilon)=0.022019.04 | 39.3 | |
| Yu et al.Adversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 38.2 | |
| BaselineAdversarial Training=No, Attack=FGSM, Perturbation Budget (epsilon)=0.022019.04 | 36.5 | |
| Pang et al.Adversarial Training=Yes, Attack=MIM, Perturbation Budget (epsilon)=0.022019.04 | 35.9 | |
| BaselineAdversarial Training=No, Attack=C&W, Perturbation Budget (c)=0.012019.04 | 35.2 | |
| AdvTrainAdversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 31 | |
| Pang et al.Adversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.022019.04 | 31 | |
| Pang et al.Adversarial Training=Yes, Attack=PGD, Perturbation Budget (epsilon)=0.022019.04 | 30.4 | |
| Pang et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.12019.04 | 25.6 | |
| AdvTrainAdversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.12019.04 | 25.4 | |
| Ross et al.Adversarial Training=Yes, Attack=MIM, Perturbation Budget (epsilon)=0.012019.04 | 24.6 | |
| Ross et al.Adversarial Training=Yes, Attack=PGD, Perturbation Budget (epsilon)=0.012019.04 | 24.5 | |
| AdvTrainAdversarial Training=Yes, Attack=PGD, Perturbation Budget (epsilon)=0.012019.04 | 24.3 | |
| AdvTrainAdversarial Training=Yes, Attack=MIM, Perturbation Budget (epsilon)=0.012019.04 | 23.9 | |
| BaselineAdversarial Training=No, Attack=MIM, Perturbation Budget (epsilon)=0.012019.04 | 23.8 | |
| BaselineAdversarial Training=No, Attack=PGD, Perturbation Budget (epsilon)=0.012019.04 | 23.4 | |
| AdvTrainAdversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 22.6 | |
| Ross et al.Adversarial Training=Yes, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 20.8 | |
| Ross et al.Adversarial Training=Yes, Attack=C&W, Perturbation Budget (c)=0.12019.04 | 19.9 | |
| BaselineAdversarial Training=No, Attack=FGSM, Perturbation Budget (epsilon)=0.042019.04 | 19.4 | |
| Ross et al.Adversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 19 | |
| BaselineAdversarial Training=No, Attack=BIM, Perturbation Budget (epsilon)=0.012019.04 | 18.5 | |
| Ross et al.Adversarial Training=Yes, Attack=MIM, Perturbation Budget (epsilon)=0.022019.04 | 9.5 | |
| AdvTrainAdversarial Training=Yes, Attack=MIM, Perturbation Budget (epsilon)=0.022019.04 | 9.3 | |
| Yu et al.Adversarial Training=Yes, Attack=PGD, Perturbation Budget (epsilon)=0.022019.04 | 8.5 | |
| AdvTrainAdversarial Training=Yes, Attack=BIM, Perturbation Budget (epsilon)=0.022019.04 | 7.8 | |
| AdvTrainAdversarial Training=Yes, Attack=PGD, Perturbation Budget (epsilon)=0.022019.04 | 7.8 | |
| BaselineAdversarial Training=No, Attack=MIM, Perturbation Budget (epsilon)=0.022019.04 | 7.4 |