Comparison of different accuracies (%) by using different methods where cross-task adversarial (Adv) and common-corruption (CC) (Hendrycks and Dietterich, 2019) robustness transferability (RT) in CIFAR-10CIFAR-100 is evaluated
| Methods | Adv-RT | CC-RT | ||||
|---|---|---|---|---|---|---|
| AA | RA | SA | CS-1 | CS-3 | CS-5 | |
| ACL* | 23.36 | 30.87 | 57.00 | 52.65 | 45.12 | 36.46 |
| ACL-AIR* | 24.23 | 31.44 | 57.24 | 53.09 | 45.96 | 37.70 |
| DynACL* | 25.07 | 32.13 | 56.43 | 53.88 | 45.77 | 37.13 |
| DynACL-AIR* | 25.63 | 32.62 | 58.10 | 54.45 | 46.90 | 38.24 |
| Ours (AT) | 26.31 | 33.61 | 58.49 | 55.62 | 47.73 | 39.26 |
| Ours-AIR (AT) | 27.68 | 35.22 | 59.75 | 57.10 | 49.05 | 40.80 |
| Methods | Adv-RT | CC-RT | ||||
|---|---|---|---|---|---|---|
| CS-1 | CS-3 | CS-5 | ||||
| ACL* | 23.36 | 30.87 | 57.00 | 52.65 | 45.12 | 36.46 |
| ACL-AIR* | 24.23 | 31.44 | 57.24 | 53.09 | 45.96 | 37.70 |
| DynACL* | 25.07 | 32.13 | 56.43 | 53.88 | 45.77 | 37.13 |
| DynACL-AIR* | 25.63 | 32.62 | 58.10 | 54.45 | 46.90 | 38.24 |
| Ours ( | 26.31 | 33.61 | 58.49 | 55.62 | 47.73 | 39.26 |
| Ours-AIR ( | ||||||
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.