Backdoor Defense on CIFAR10 (train)
0.31ASRANP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ANPBackdoor Attack=WaNet, Data Type=In-distribution Labeled2022.11 | 0.31 | 90.61 | |
| ANPBackdoor Attack=SIG, Data Type=In-distribution Labeled2022.11 | 0.41 | 92.09 | |
| ANPBackdoor Attack=IAB, Data Type=In-distribution Labeled2022.11 | 0.6 | 85.37 | |
| ANPBackdoor Attack=Blended, Data Type=In-distribution Labeled2022.11 | 0.87 | 92.85 | |
| BCUBackdoor Attack=SIG, Data Type=In-distribution Unlabeled, Unlabeled Samples=25002022.11 | 0.91 | 92.58 | |
| FinetuningBackdoor Attack=WaNet, Data Type=In-distribution Labeled2022.11 | 0.98 | 92.34 | |
| NADBackdoor Attack=WaNet, Data Type=In-distribution Labeled2022.11 | 1.03 | 92.22 | |
| MCRBackdoor Attack=WaNet, Data Type=In-distribution Labeled, t=0.32022.11 | 1.14 | 91.08 | |
| BCUBackdoor Attack=SIG, Data Type=In-distribution Unlabeled, Unlabeled Samples=50002022.11 | 1.18 | 93.14 | |
| MCRBackdoor Attack=IAB, Data Type=In-distribution Labeled, t=0.32022.11 | 1.35 | 85.29 | |
| BCUBackdoor Attack=LC, Data Type=In-distribution Unlabeled, Unlabeled Samples=50002022.11 | 1.4 | 93.66 | |
| ANPBackdoor Attack=Mean, Data Type=In-distribution Labeled2022.11 | 1.56 | 90.14 | |
| MCRBackdoor Attack=Badnets, Data Type=In-distribution Labeled, t=0.32022.11 | 1.68 | 86.41 | |
| I-BAUBackdoor Attack=WaNet, Data Type=In-distribution Labeled2022.11 | 1.73 | 91.62 | |
| BCUBackdoor Attack=LC, Data Type=In-distribution Unlabeled, Unlabeled Samples=25002022.11 | 1.81 | 93.17 | |
| NADBackdoor Attack=SIG, Data Type=In-distribution Labeled2022.11 | 1.88 | 92.95 | |
| BCUBackdoor Attack=IAB, Data Type=In-distribution Unlabeled, Unlabeled Samples=50002022.11 | 1.9 | 86.85 | |
| BCUBackdoor Attack=IAB, Data Type=In-distribution Unlabeled, Unlabeled Samples=25002022.11 | 1.96 | 86.42 | |
| NADBackdoor Attack=IAB, Data Type=In-distribution Labeled2022.11 | 2.17 | 86.76 | |
| MCRBackdoor Attack=SIG, Data Type=In-distribution Labeled, t=0.32022.11 | 2.33 | 87.69 | |
| Fine-pruningBackdoor Attack=IAB, Data Type=In-distribution Labeled2022.11 | 2.45 | 86.89 | |
| ANPBackdoor Attack=Badnets, Data Type=In-distribution Labeled2022.11 | 2.56 | 88.58 | |
| BCUBackdoor Attack=Badnets, Data Type=In-distribution Unlabeled, Unlabeled Samples=25002022.11 | 3 | 92.15 | |
| BCUBackdoor Attack=Badnets, Data Type=In-distribution Unlabeled, Unlabeled Samples=50002022.11 | 3 | 92.75 | |
| MCRBackdoor Attack=Mean, Data Type=In-distribution Labeled, t=0.32022.11 | 3.04 | 87.69 | |
| BCUBackdoor Attack=Mean, Data Type=In-distribution Unlabeled, Unlabeled Samples=25002022.11 | 3.74 | 91.59 | |
| ANPBackdoor Attack=LC, Data Type=In-distribution Labeled2022.11 | 4.62 | 91.3 | |
| NADBackdoor Attack=Badnets, Data Type=In-distribution Labeled2022.11 | 4.67 | 92.35 | |
| BCUBackdoor Attack=Mean, Data Type=In-distribution Unlabeled, Unlabeled Samples=50002022.11 | 4.87 | 92.11 | |
| BCUBackdoor Attack=Blended, Data Type=In-distribution Unlabeled, Unlabeled Samples=25002022.11 | 4.9 | 93.16 | |
| NADBackdoor Attack=Blended, Data Type=In-distribution Labeled2022.11 | 5.06 | 93.24 | |
| BCUBackdoor Attack=Blended, Data Type=In-distribution Unlabeled, Unlabeled Samples=50002022.11 | 5.1 | 93.65 | |
| FinetuningBackdoor Attack=Blended, Data Type=In-distribution Labeled2022.11 | 5.2 | 93.44 | |
| MCRBackdoor Attack=LC, Data Type=In-distribution Labeled, t=0.32022.11 | 5.33 | 88.18 | |
| FinetuningBackdoor Attack=SIG, Data Type=In-distribution Labeled2022.11 | 5.41 | 93.16 | |
| Fine-pruningBackdoor Attack=SIG, Data Type=In-distribution Labeled2022.11 | 5.66 | 93.55 | |
| I-BAUBackdoor Attack=Blended, Data Type=In-distribution Labeled2022.11 | 6.19 | 92.71 | |
| MCRBackdoor Attack=Blended, Data Type=In-distribution Labeled, t=0.32022.11 | 6.39 | 87.51 | |
| I-BAUBackdoor Attack=IAB, Data Type=In-distribution Labeled2022.11 | 7.57 | 85.64 | |
| FinetuningBackdoor Attack=IAB, Data Type=In-distribution Labeled2022.11 | 9.46 | 86.91 | |
| FinetuningBackdoor Attack=Badnets, Data Type=In-distribution Labeled2022.11 | 9.7 | 92.55 | |
| BCUBackdoor Attack=WaNet, Data Type=In-distribution Unlabeled, Unlabeled Samples=25002022.11 | 9.86 | 92.05 | |
| I-BAUBackdoor Attack=Badnets, Data Type=In-distribution Labeled2022.11 | 10.16 | 91.98 | |
| I-BAUBackdoor Attack=Mean, Data Type=In-distribution Labeled2022.11 | 10.47 | 91.19 | |
| NADBackdoor Attack=Mean, Data Type=In-distribution Labeled2022.11 | 11.26 | 91.82 | |
| Fine-pruningBackdoor Attack=WaNet, Data Type=In-distribution Labeled2022.11 | 13.99 | 92.92 | |
| I-BAUBackdoor Attack=SIG, Data Type=In-distribution Labeled2022.11 | 15.76 | 92.45 | |
| BCUBackdoor Attack=WaNet, Data Type=In-distribution Unlabeled, Unlabeled Samples=50002022.11 | 16.67 | 92.64 | |
| Fine-pruningBackdoor Attack=Blended, Data Type=In-distribution Labeled2022.11 | 20.62 | 93.7 | |
| FinetuningBackdoor Attack=Mean, Data Type=In-distribution Labeled2022.11 | 21.31 | 91.98 | |
| I-BAUBackdoor Attack=LC, Data Type=In-distribution Labeled2022.11 | 21.41 | 92.72 | |
| Fine-pruningBackdoor Attack=Mean, Data Type=In-distribution Labeled2022.11 | 22.55 | 92.25 | |
| Fine-pruningBackdoor Attack=Badnets, Data Type=In-distribution Labeled2022.11 | 32.36 | 92.57 | |
| NADBackdoor Attack=LC, Data Type=In-distribution Labeled2022.11 | 52.74 | 93.38 | |
| Fine-pruningBackdoor Attack=LC, Data Type=In-distribution Labeled2022.11 | 60.23 | 93.88 | |
| Original (No Defense)Backdoor Attack=IAB2022.11 | 91.35 | 87.46 | |
| Original (No Defense)Backdoor Attack=SIG2022.11 | 95.09 | 93.71 | |
| FinetuningBackdoor Attack=LC, Data Type=In-distribution Labeled2022.11 | 97.14 | 93.49 | |
| Original (No Defense)Backdoor Attack=WaNet2022.11 | 97.15 | 93.53 | |
| Original (No Defense)Backdoor Attack=Mean2022.11 | 97.18 | 92.74 | |
| Original (No Defense)Backdoor Attack=LC2022.11 | 99.55 | 94.51 | |
| Original (No Defense)Backdoor Attack=Badnets2022.11 | 99.93 | 92.76 | |
| Original (No Defense)Backdoor Attack=Blended2022.11 | 100 | 94.48 |