Backdoor Defense on CIFAR-10 1.0 (test)
94.04BadNet C-AccOriginal Backdoored Model
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Original Backdoored ModelArchitecture=ResNet18, Tuning Data Ratio=5%, Evaluation Mode=O-Backdoor2024.10 | 94.04 | 99.99 | 94.77 | 100 | 94.6 | 96.38 | 94.86 | 99.99 | 94.57 | 99.09 | |
| Original Backdoored ModelArchitecture=ResNet50, Tuning Data Ratio=5%, Evaluation Mode=O-Backdoor2024.10 | 93.81 | 99.92 | 94.53 | 100 | 93.65 | 97.7 | 94.57 | 99.9 | 94.14 | 99.38 | |
| Original Backdoored ModelArchitecture=ResNet18, Tuning Data Ratio=10%, Evaluation Mode=O-Backdoor2024.10 | 93.73 | 100 | 94.28 | 100 | 94.31 | 98.71 | 85.8 | 100 | 92.03 | 99.68 | |
| EPArchitecture=ResNet18, Tuning Data Ratio=5%, Evaluation Mode=O-Robustness2024.10 | 93.57 | 0.94 | 93.47 | 3.37 | 93.39 | 0.53 | 93.64 | 0.52 | 93.52 | 1.34 | |
| EPArchitecture=ResNet50, Tuning Data Ratio=5%, Evaluation Mode=O-Robustness2024.10 | 92.84 | 0.98 | 92.1 | 1.12 | 91.84 | 0.43 | 92.91 | 0.41 | 92.42 | 0.74 | |
| EPArchitecture=ResNet18, Tuning Data Ratio=10%, Evaluation Mode=O-Robustness2024.10 | 92.78 | 0.93 | 93.34 | 1.52 | 93.1 | 0.71 | 92.12 | 0.16 | 92.84 | 0.83 | |
| PAMArchitecture=ResNet50, Tuning Data Ratio=5%, Evaluation Mode=O-Robustness2024.10 | 92.58 | 0.94 | 92.59 | 0.24 | 92.36 | 1.62 | 91.99 | 2.14 | 92.38 | 1.23 | |
| PAMArchitecture=ResNet18, Tuning Data Ratio=10%, Evaluation Mode=O-Robustness2024.10 | 92.43 | 1.73 | 92.63 | 0.22 | 92.89 | 1.37 | 91.06 | 2.31 | 92.25 | 1.41 | |
| EPArchitecture=ResNet50, Tuning Data Ratio=5%, Evaluation Mode=P-Robustness2024.10 | 92.33 | 1.48 | 91.36 | 3.51 | 89.26 | 6.28 | 92.33 | 1.71 | 91.32 | 3.25 | |
| EPArchitecture=ResNet18, Tuning Data Ratio=5%, Evaluation Mode=P-Robustness2024.10 | 92.27 | 1.08 | 93.36 | 4.93 | 92.05 | 2.47 | 93.69 | 0.41 | 92.84 | 2.22 | |
| EPArchitecture=ResNet18, Tuning Data Ratio=10%, Evaluation Mode=P-Robustness2024.10 | 92.17 | 1.58 | 93.06 | 4.78 | 92.04 | 3.7 | 91.79 | 2.03 | 92.27 | 3.02 | |
| PAMArchitecture=ResNet18, Tuning Data Ratio=5%, Evaluation Mode=O-Robustness2024.10 | 92.11 | 1.14 | 93.34 | 1.67 | 92.96 | 1.24 | 92.32 | 4.92 | 92.68 | 2.24 | |
| PAMArchitecture=ResNet18, Tuning Data Ratio=10%, Evaluation Mode=P-Robustness2024.10 | 91.77 | 1.47 | 91.85 | 7.44 | 92.01 | 2.63 | 90.98 | 3.37 | 91.65 | 3.73 | |
| PAMArchitecture=ResNet18, Tuning Data Ratio=5%, Evaluation Mode=P-Robustness2024.10 | 91.66 | 3.9 | 93.38 | 2.69 | 92.2 | 3.31 | 92.15 | 8.31 | 92.35 | 4.55 | |
| PAMArchitecture=ResNet50, Tuning Data Ratio=5%, Evaluation Mode=P-Robustness2024.10 | 91.49 | 1.46 | 92.75 | 0.64 | 92.32 | 14.23 | 90.6 | 3.54 | 91.79 | 4.97 | |
| Original Backdoored ModelArchitecture=DenseNet161, Tuning Data Ratio=5%, Evaluation Mode=O-Backdoor2024.10 | 89.85 | 100 | 89.59 | 98.72 | 88.83 | 86.75 | 90.13 | 99.8 | 89.6 | 96.32 | |
| PAMArchitecture=DenseNet161, Tuning Data Ratio=5%, Evaluation Mode=O-Robustness2024.10 | 88.7 | 1.22 | 87.02 | 1.08 | 87.62 | 1.61 | 87.7 | 1.81 | 87.76 | 1.43 | |
| EPArchitecture=DenseNet161, Tuning Data Ratio=5%, Evaluation Mode=O-Robustness2024.10 | 88.58 | 1.52 | 88 | 3.21 | 88.17 | 0.69 | 88.59 | 0.56 | 88.34 | 1.5 | |
| EPArchitecture=DenseNet161, Tuning Data Ratio=5%, Evaluation Mode=P-Robustness2024.10 | 88.03 | 2.77 | 87.13 | 3.54 | 86.33 | 1.47 | 88.94 | 0.95 | 87.61 | 2.18 | |
| PAMArchitecture=DenseNet161, Tuning Data Ratio=5%, Evaluation Mode=P-Robustness2024.10 | 87.03 | 3.04 | 86.49 | 1.94 | 85.89 | 3.72 | 86.44 | 9.39 | 86.42 | 4.52 |