Machine Unlearning on CIFAR-100 (train)
99.99Accuracy ($D_f$)Original
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OriginalBackbone=ResNet-182026.03 | 99.99 | 100 | 100 | |
| OriginalBackbone=ViT-B2026.03 | 99.93 | 99.9 | 100 | |
| FTBackbone=ResNet-182026.03 | 76.67 | 99.47 | 99.47 | |
| FTBackbone=ViT-B2026.03 | 76.67 | 99.47 | 99.47 | |
| GABackbone=ResNet-182026.03 | 70.53 | 72.75 | 91.33 | |
| ℓ1-sparseBackbone=ViT-B2026.03 | 62 | 98.41 | 98.81 | |
| ℓ1-sparseBackbone=ResNet-182026.03 | 55.93 | 98.48 | 96.92 | |
| Retrain2026.03 | 38.4 | 99.98 | 99.99 | |
| SSDBackbone=ResNet-182026.03 | 37.4 | 43.75 | 76.02 | |
| MunbaBackbone=ResNet-182026.03 | 33.8 | 92.17 | 92.68 | |
| GDRBackbone=ResNet-182026.03 | 4.87 | 31.92 | 96.1 | |
| SCRUBBackbone=ResNet-182026.03 | 4.47 | 58.65 | 82.67 | |
| SalUnBackbone=ResNet-182026.03 | 3.2 | 52.27 | 86.35 | |
| SCRUBBackbone=ViT-B2026.03 | 1.93 | 99.98 | 99.66 | |
| Two-Stage Machine Unlearning2026.03 | 0 | 98.17 | 98.44 | |
| DELETEBackbone=ResNet-182026.03 | 0 | 3.57 | 98.37 | |
| Our methodBackbone=ResNet-182026.03 | 0 | 98.17 | 98.44 | |
| GABackbone=ViT-B2026.03 | 0 | 16.41 | 90.64 | |
| Our methodBackbone=ViT-B2026.03 | 0 | 98.5 | 98.87 |