Backdoor Defense on CIFAR-10 IAB-all attack clean data 10% (train)
94.1AccuracyBefore
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BeforeBackbone=ResNet-182021.10 | 94.1 | 92.88 | |
| ANPBackbone=ResNet-182021.10 | 93.56 | 0.79 | |
| MCRBackbone=ResNet-18, Hyperparameter=0.32021.10 | 88.53 | 17.17 | |
| Fine-tuning (FT)Backbone=ResNet-18, Hyperparameter=0.012021.10 | 86.54 | 1.91 | |
| Fine-tuning (FT)Backbone=ResNet-18, Hyperparameter=0.022021.10 | 85.62 | 2.15 | |
| Fine-pruning (FP)Backbone=ResNet-182021.10 | 85.26 | 6.6 |