Image Classification on CIFAR-10 (test) (Early Exit)
46.53Accuracy (Exit 1)NEO-KD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| NEO-KDAnytime prediction setup=true, Attack=Max-average attack, Perturbation degree (epsilon)=8/255, PGD attack=50 steps2023.11 | 46.53 | 47.65 | 50.71 | 48.3 | |
| Adv. w/o DistillAnytime prediction setup=true, Attack=Max-average attack, Perturbation degree (epsilon)=8/255, PGD attack=50 steps2023.11 | 44.77 | 46.1 | 46.86 | 45.91 | |
| SKDAnytime prediction setup=true, Attack=Max-average attack, Perturbation degree (epsilon)=8/255, PGD attack=50 steps2023.11 | 44.75 | 44.54 | 44.79 | 44.69 | |
| ARDAnytime prediction setup=true, Attack=Max-average attack, Perturbation degree (epsilon)=8/255, PGD attack=50 steps2023.11 | 44.5 | 45.85 | 51.82 | 47.39 | |
| LWAnytime prediction setup=true, Attack=Max-average attack, Perturbation degree (epsilon)=8/255, PGD attack=50 steps2023.11 | 37.38 | 35.39 | 34.85 | 35.87 |