Certified Accuracy on CIFAR-10 (test)
86.61Certified Accuracy (r=0.0)Carlini et al.
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Carlini et al.sigma=0.252025.12 | 86.61 | 73.9 | 57.02 | 35.3 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | |
| Multiscale + finetuningsigma={0.25, 0.5, 1.0}2025.12 | 76.51 | 54.78 | 39.15 | 28.46 | 21.33 | 15.95 | 11.4 | 7.91 | 5.31 | 3.63 | 2.34 | |
| Carlini et al.sigma=0.52025.12 | 73.49 | 62.23 | 49.46 | 38.2 | 28.58 | 19.54 | 11.22 | 0 | 0 | 0 | 0 | |
| Dual RS + finetuningsigma={0.25, 0.5, 1.0}2025.12 | 71.12 | 58.93 | 46.55 | 35.42 | 25.75 | 18.35 | 12.22 | 8.2 | 6.18 | 4.35 | 2.96 | |
| Dual RSsigma={0.25, 0.5, 1.0}2025.12 | 68.91 | 55.45 | 41.62 | 29.48 | 20.26 | 13.29 | 7.95 | 4.7 | 3.45 | 2.4 | 1.7 | |
| MDEQ-30ABackbone=MDEQ-30A, sigma=0.52024.11 | 67 | 55 | 45 | 33 | 23 | 16 | 12 | — | — | — | — | |
| SRS-MDEQ-3ABackbone=MDEQ-30A, sigma=0.5, mode=Serialized Randomized Smoothing2024.11 | 66 | 54 | 45 | 33 | 23 | 16 | 11 | — | — | — | — | |
| ResNet-110Backbone=ResNet-110, sigma=0.52024.11 | 65 | 54 | 41 | 32 | 23 | 15 | 9 | — | — | — | — | |
| Carlini et al.sigma=1.02025.12 | 47.98 | 39.85 | 32.17 | 25.16 | 19.34 | 14.49 | 10.3 | 7.32 | 4.89 | 3.35 | 2.15 |