Trojan Defense on MNIST
99.22Benign AccuracyUndefended
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Undefendedsetup=Federated Learning2022.02 | 99.22 | 99.15 | — | — | — | — | |
| UndefendedNetwork=LeNet5, Attack=WaNet2022.10 | 99.22 | 94.52 | — | — | — | — | |
| FeatureRENetwork=LeNet5, Attack=WaNet2022.10 | 99.2 | 0.63 | — | — | — | — | |
| NONEsetup=Federated Learning2022.02 | 98.6 | 0.13 | — | — | — | — | |
| I-BAUNetwork=LeNet5, Attack=WaNet2022.10 | 94.87 | 0.22 | — | — | — | — | |
| FeatureRENetwork=LeNet5, Attack=WaNet2022.10 | 90 | — | 9 | 1 | 1 | 9 | |
| ABSNetwork=LeNet5, Attack=WaNet2022.10 | 75 | — | 7 | 2 | 3 | 8 | |
| K-armNetwork=LeNet5, Attack=WaNet2022.10 | 75 | — | 5 | 0 | 5 | 10 | |
| DeepInspectNetwork=LeNet5, Attack=WaNet2022.10 | 70 | — | 4 | 0 | 6 | 10 | |
| NADNetwork=LeNet5, Attack=WaNet2022.10 | 65.87 | 48.21 | — | — | — | — | |
| TABORNetwork=LeNet5, Attack=WaNet2022.10 | 55 | — | 3 | 2 | 7 | 8 |