Backdoor Defense on CIFAR-10 clean data SIG attack 10% (train)
93.65AccuracyANP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ANPBackbone=ResNet-182021.10 | 93.65 | 0.28 | |
| BeforeBackbone=ResNet-182021.10 | 93.64 | 94.26 | |
| Fine-pruning (FP)Backbone=ResNet-182021.10 | 87.47 | 0.62 | |
| Fine-tuning (FT)Backbone=ResNet-18, Hyperparameter=0.012021.10 | 86.95 | 0.86 | |
| MCRBackbone=ResNet-18, Hyperparameter=0.32021.10 | 85.88 | 0.52 | |
| Fine-tuning (FT)Backbone=ResNet-18, Hyperparameter=0.022021.10 | 85.8 | 0.92 |