Attributional Robustness on CIFAR-10 (test)
1.81Sensitivity ScoreInput-Grad Spatial
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Input-Grad SpatialBackbone=ResNet182024.11 | 1.81 | — | — | — | — | — | — | |
| Input-Grad SpatialBackbone=VGG162024.11 | 1.671 | — | — | — | — | — | — | |
| IG SpatialBackbone=ResNet182024.11 | 1.465 | — | — | — | — | — | — | |
| IG SpatialBackbone=VGG162024.11 | 1.187 | — | — | — | — | — | — | |
| Input-Grad FFTBackbone=ResNet182024.11 | 1.174 | — | — | — | — | — | — | |
| Input-Grad IFFTBackbone=ResNet182024.11 | 1.12 | — | — | — | — | — | — | |
| Input-Grad FFTBackbone=VGG162024.11 | 1.1 | — | — | — | — | — | — | |
| IG FFTBackbone=ResNet182024.11 | 1.029 | — | — | — | — | — | — | |
| IG IFFTBackbone=ResNet182024.11 | 1.029 | — | — | — | — | — | — | |
| Input-Grad IFFTBackbone=VGG162024.11 | 1.009 | — | — | — | — | — | — | |
| Smooth-Grad SpatialBackbone=ResNet182024.11 | 0.771 | — | — | — | — | — | — | |
| IG FFTBackbone=VGG162024.11 | 0.756 | — | — | — | — | — | — | |
| IG IFFTBackbone=VGG162024.11 | 0.756 | — | — | — | — | — | — | |
| Smooth-Grad SpatialBackbone=VGG162024.11 | 0.728 | — | — | — | — | — | — | |
| FFCBackbone=ResNet182024.11 | 0.5339 | — | — | — | — | — | — | |
| Smooth-Grad FFTBackbone=ResNet182024.11 | 0.42 | — | — | — | — | — | — | |
| Smooth-Grad FFTBackbone=VGG162024.11 | 0.406 | — | — | — | — | — | — | |
| Full-Grad FFTBackbone=VGG162024.11 | 0.406 | — | — | — | — | — | — | |
| Grad-CAM SpatialBackbone=ResNet182024.11 | 0.295 | — | — | — | — | — | — | |
| Grad-CAM FFTBackbone=ResNet182024.11 | 0.284 | — | — | — | — | — | — | |
| Grad-CAM IFFTBackbone=ResNet182024.11 | 0.275 | — | — | — | — | — | — | |
| Smooth-Grad IFFTBackbone=ResNet182024.11 | 0.275 | — | — | — | — | — | — | |
| Smooth-Grad IFFTBackbone=VGG162024.11 | 0.26 | — | — | — | — | — | — | |
| Full-Grad IFFTBackbone=VGG162024.11 | 0.26 | — | — | — | — | — | — | |
| Grad-CAM SpatialBackbone=VGG162024.11 | 0.171 | — | — | — | — | — | — | |
| Grad-CAM FFTBackbone=VGG162024.11 | 0.17 | — | — | — | — | — | — | |
| Grad-CAM IFFTBackbone=VGG162024.11 | 0.166 | — | — | — | — | — | — | |
| Full-Grad SpatialBackbone=ResNet182024.11 | 0.127 | — | — | — | — | — | — | |
| Full-Grad IFFTBackbone=ResNet182024.11 | 0.1173 | — | — | — | — | — | — | |
| Full-Grad FFTBackbone=ResNet182024.11 | 0.117 | — | — | — | — | — | — | |
| FFCBackbone=VGG162024.11 | 0.1061 | — | — | — | — | — | — | |
| Full-Grad SpatialBackbone=VGG162024.11 | 0.0757 | — | — | — | — | — | — | |
| AdvAATAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 72.11 | 0.5484 | |
| ARTTraining=Attributional Robust Training (ART)2019.11 | — | 0.7607 | 0.7031 | 0.4831 | 0.6235 | — | — | |
| ARTAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 70.44 | 0.6875 | |
| ATAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 72.21 | 0.5578 | |
| AT+IGRAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 73.37 | 0.5775 | |
| IG-NORMAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 74.49 | 0.5811 | |
| IG-SUM-NORMAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 78.7 | 0.6901 | |
| MARTAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 76.11 | 0.6192 | |
| MART+IGRAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 76.56 | 0.6328 | |
| NaturalTraining=Standard2019.11 | — | 0.1372 | 0.095 | 0.045 | 0.1652 | — | — | |
| PGD-10Training=Adversarial Training2019.11 | — | 0.548 | 0.5406 | 0.4505 | 0.598 | — | — | |
| SSRAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 71.2 | 0.5498 | |
| StandardAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 46.71 | 0.1662 | |
| TRADESAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 78.28 | 0.6903 | |
| TRADES+AdvAATAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 71.3 | 0.5239 | |
| TRADES+IGRAttack Method=IFIA, Attack Steps=200, epsilon=8/255, k=1000, Backbone=ResNet-182022.05 | — | — | — | — | — | 80.26 | 0.694 |