Certified Robustness on CIFAR-10 (L2 Accuracy Profile)
85.4Certified Accuracy (L2, r=0.0)LBS
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LBSsigma=0.5, N=100,0002026.04 | 85.4 | 84.4 | 83.4 | 81.1 | 75.2 | 0.7 | |
| UCANTraining Augmentation=noise-augmented (w/), sigma=0.5, N=100,0002026.04 | 83.3 | 81.7 | 80.6 | 78.4 | 74 | 9.4 | |
| EnsembleTraining Augmentation=noise-augmented (w/), sigma=0.5, N=100,0002026.04 | 68.8 | 64.7 | 60.1 | 48.4 | 29.5 | 9.2 | |
| ASTraining Augmentation=noise-augmented (w/), sigma=0.5, N=100,0002026.04 | 66.4 | 63.8 | 59.3 | 48.1 | 30.4 | 0.4 | |
| ADRETraining Augmentation=noise-augmented (w/), sigma=0.5, N=100,0002026.04 | 65.2 | 61 | 56.4 | 46.7 | 28.1 | 9.6 | |
| Betting CSTraining Augmentation=noise-augmented (w/), sigma=0.5, N=100,0002026.04 | 63.6 | 58.8 | 51.2 | 40.9 | 22 | 1.2 | |
| RSTraining Augmentation=noise-augmented (w/), sigma=0.5, N=100,0002026.04 | 63.4 | 58.4 | 51.8 | 41.4 | 21.8 | 9.8 | |
| ISSTraining Augmentation=noise-augmented (w/), sigma=0.5, N=100,0002026.04 | 61.2 | 56.6 | 49.9 | 39.3 | 20 | 6.6 | |
| ASTraining Augmentation=standard (w/o), sigma=0.5, N=100,0002026.04 | 13 | 11.6 | 9.8 | 6 | 0 | 0.4 | |
| ADRETraining Augmentation=standard (w/o), sigma=0.5, N=100,0002026.04 | 12.6 | 10 | 8.4 | 4.2 | 0 | 9.6 | |
| UCANTraining Augmentation=standard (w/o), sigma=0.5, N=100,0002026.04 | 10.8 | 9.2 | 7.6 | 4 | 0 | 9.4 | |
| Betting CSTraining Augmentation=standard (w/o), sigma=0.5, N=100,0002026.04 | 9.1 | 8 | 6.2 | 4.7 | 0 | 1.2 | |
| RSTraining Augmentation=standard (w/o), sigma=0.5, N=100,0002026.04 | 8.4 | 6.2 | 4.8 | 0.8 | 0 | 9.8 | |
| ISSTraining Augmentation=standard (w/o), sigma=0.5, N=100,0002026.04 | 6.8 | 4.4 | 0.8 | 0.4 | 0 | 6.6 |