Adversarial Robust Image Classification on CIFAR-10 (test) (L1/L2/Linf Accuracy)
84.2Clean AccuracyAVG
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AVGBackbone=PreAct-ResNet-18, Fine-tuning epochs=3, Seeds=5, Pre-trained checkpoint=RN-18-l_inf-AT2024.02 | 84.2 | 43.3 | 68.4 | 46.9 | 40.6 | |
| RN-18-l_inf-ATBackbone=PreAct-ResNet-18, Status=Starting model baseline2024.02 | 83.7 | 48.1 | 59.8 | 7.7 | 38.5 | |
| SATBackbone=PreAct-ResNet-18, Fine-tuning epochs=3, Seeds=5, Pre-trained checkpoint=RN-18-l_inf-AT2024.02 | 83.5 | 43.5 | 68 | 47.4 | 41 | |
| E-ATBackbone=PreAct-ResNet-18, Fine-tuning epochs=3, Seeds=5, Pre-trained checkpoint=RN-18-l_inf-AT2024.02 | 82.7 | 44.3 | 68.1 | 48.7 | 42.2 | |
| MAXBackbone=PreAct-ResNet-18, Fine-tuning epochs=3, Seeds=5, Pre-trained checkpoint=RN-18-l_inf-AT2024.02 | 82.2 | 45.2 | 67 | 46.1 | 42.2 | |
| MSDBackbone=PreAct-ResNet-18, Fine-tuning epochs=3, Seeds=5, Pre-trained checkpoint=RN-18-l_inf-AT2024.02 | 82.2 | 44.9 | 67.1 | 47.2 | 42.6 | |
| RAMPBackbone=PreAct-ResNet-18, Fine-tuning epochs=3, Seeds=5, Pre-trained checkpoint=RN-18-l_inf-AT, lambda=0.52024.02 | 81.5 | 45.5 | 66.4 | 47 | 42.9 | |
| RAMPBackbone=PreAct-ResNet-18, Fine-tuning epochs=3, Seeds=5, Pre-trained checkpoint=RN-18-l_inf-AT, lambda=1.52024.02 | 81.1 | 45.4 | 66.1 | 47.2 | 43.1 |