Backdoor Defense on CIFAR-10 10% clean data (CLB attack, train)
93.78AccuracyBefore
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BeforeBackbone=ResNet-182021.10 | 93.78 | 99.94 | |
| ANPBackbone=ResNet-182021.10 | 93.37 | 3.65 | |
| MCRBackbone=ResNet-18, Hyperparameter=0.32021.10 | 89.78 | 3.77 | |
| Fine-tuning (FT)Backbone=ResNet-18, Hyperparameter=0.012021.10 | 87.12 | 1.56 | |
| Fine-pruning (FP)Backbone=ResNet-182021.10 | 86.89 | 1.97 | |
| Fine-tuning (FT)Backbone=ResNet-18, Hyperparameter=0.022021.10 | 84.7 | 2.27 |