Image Classification on CIFAR10 (Adversarial Robustness Metrics)
94.3Clean AccuracyMarksman
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MarksmanPoisoning Rate=10%2022.10 | 94.3 | 0.5 | 100 | |
| MarksmanPoisoning rate=50%2022.10 | 94.1 | 0.7 | 100 | |
| MarksmanPost-training Defense Method=Adversarial Neuron Pruning (ANP)2022.10 | 93.8 | — | 64.7 | |
| RefoolMTPoisoning Rate=10%2022.10 | 93.4 | 0.6 | 73 | |
| PatchMTPoisoning Rate=10%2022.10 | 93.3 | 0.7 | 48.7 | |
| WaNetMTPoisoning rate=50%2022.10 | 92 | 2 | 99.9 | |
| RefoolMTPoisoning rate=50%2022.10 | 91 | 3 | 98.4 | |
| WaNetMTPoisoning Rate=10%2022.10 | 89.4 | 4.6 | 30.8 | |
| PatchMTPoisoning rate=50%2022.10 | 88.2 | 5.8 | 99 | |
| MarksmanPost-training Defense Method=Neural Attention Distillation (NAD)2022.10 | 36.5 | — | 6.9 |