Backdoor Defense on CIFAR-10 (1% clean data (500 images))
93.73ACC (Badnets)No Defense
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| No DefenseBackbone=ResNet-18, Clean Data Images=5002021.10 | 93.73 | 94.82 | 93.89 | 94.1 | 93.78 | 93.64 | 99.97 | 100 | 98.49 | 92.88 | 99.94 | 94.26 | |
| Fine-pruning (FP)Backbone=ResNet-18, Clean Data Images=5002021.10 | 92.18 | 93.4 | 91.57 | 92.28 | 91.91 | 91.64 | 5.34 | 65.39 | 20.73 | 32.36 | 3.4 | 0.32 | |
| Fine-tuning (FT)Backbone=ResNet-18, ratio=0.01, Clean Data Images=5002021.10 | 90.48 | 92.12 | 88.68 | 89.06 | 91.26 | 91.19 | 11.7 | 47.17 | 0.99 | 1.36 | 12.51 | 0.4 | |
| Adversarial Neuron Pruning (ANP)Backbone=ResNet-18, Clean Data Images=5002021.10 | 90.2 | 93.44 | 92.62 | 92.79 | 92.67 | 93.4 | 0.45 | 0.46 | 0.88 | 0.86 | 3.98 | 0.28 | |
| Fine-tuning (FT)Backbone=ResNet-18, ratio=0.02, Clean Data Images=5002021.10 | 87.23 | 88.98 | 84.85 | 83.77 | 88.25 | 88.63 | 2.95 | 10.2 | 1.7 | 1.83 | 1.17 | 0.39 | |
| Model Cleanse and Repair (MCR)Backbone=ResNet-18, parameter=0.3, Clean Data Images=5002021.10 | 85.95 | 88.26 | 86.3 | 84.53 | 86.87 | 85.88 | 5.7 | 13.57 | 30.23 | 35.17 | 12.77 | 0.52 |