Machine Unlearning on CIFAR-10 Random Forget (10%)
100Unlearn AccuracyRL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RLAccess to=Only DF2025.12 | 100 | 100 | 94.45 | 61.85 | 4.31 | |
| BSAccess to=Only DF2025.12 | 100 | 100 | 94.48 | 61.41 | 4.2 | |
| SalUnAccess to=Only DF2025.12 | 100 | 100 | 94.47 | 61.09 | 4.11 | |
| BSBackbone=ResNet-18, Access to DR=true2025.12 | 99.14 | 99.89 | 93.04 | 57.85 | 3.48 | |
| GABackbone=ResNet-18, Access to DR=true2025.12 | 98.94 | 99.22 | 93.39 | 60.96 | 4.28 | |
| SalUnBackbone=ResNet-18, Access to DR=true2025.12 | 96.25 | 98.14 | 93.06 | 50.88 | 1.44 | |
| RLBackbone=ResNet-18, Access to DR=true2025.12 | 95.54 | 97.47 | 92.17 | 51.33 | 1.74 | |
| AMUNBackbone=ResNet-18, Access to DR=true2025.12 | 95.45 | 99.57 | 93.45 | 50.18 | 0.62 | |
| FTBackbone=ResNet-18, Access to DR=true2025.12 | 95.16 | 96.64 | 92.21 | 52.08 | 2.06 | |
| Amun+SalUnBackbone=ResNet-18, Access to DR=true2025.12 | 95.02 | 99.58 | 93.29 | 50.72 | 0.68 | |
| RetrainBackbone=ResNet-18, Access to DR=true2025.12 | 94.49 | 100 | 94.33 | 50 | 0 | |
| RetrainAccess to=Only DF2025.12 | 94.49 | 100 | 94.33 | 50 | 0 | |
| l1-SparseBackbone=ResNet-18, Access to DR=true2025.12 | 94.29 | 95.63 | 91.55 | 51.21 | 2.16 | |
| AMUNAccess to=Only DF2025.12 | 94.28 | 97.47 | 91.67 | 52.24 | 1.94 | |
| AMUN+SalUnAccess to=Only DF2025.12 | 94.19 | 97.71 | 91.79 | 51.93 | 1.77 | |
| GAAccess to=Only DF2025.12 | 4.77 | 5.07 | 5.09 | 49.78 | 68.53 |