Image Classification on CIFAR100 (test) (Gradient Norms and Accuracy)
22.04Ex ||∇f(x)||^2LCNNs
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LCNNsBackbone=ResNet-18, Regularizer=LCNN2022.06 | 22.04 | 1,143.62 | 69.5 | 77.3 | |
| StandardBackbone=ResNet-18, Regularizer=None2022.06 | 19.66 | 6,061.96 | 270.89 | 77.42 | |
| Softplus + Wt. DecayBackbone=ResNet-18, Regularizer=Softplus and Weight Decay2022.06 | 18.08 | 1,052.84 | 70.39 | 77.44 | |
| LCNNs + GradRegBackbone=ResNet-18, Regularizer=LCNN + Gradient Norm2022.06 | 9.87 | 154.36 | 25.3 | 77.29 | |
| GradRegBackbone=ResNet-18, Regularizer=Gradient Norm2022.06 | 8.86 | 776.56 | 89.47 | 77.2 | |
| CUREBackbone=ResNet-18, Regularizer=Curvature Regularization2022.06 | 8.86 | 979.45 | 116.31 | 76.48 | |
| Adversarial TrainingBackbone=ResNet-18, Regularizer=l2 PGD2022.06 | 7.99 | 501.43 | 63.79 | 76.96 |