Machine Unlearning on CIFAR-10 Random Forget 50% (train)
100Unlearn AccRL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RLAccess=Only DF2025.12 | 100 | 100 | 94.54 | 0.4 | 6.75 | |
| GAAccess=Only DF2025.12 | 100 | 100 | 94.57 | 0.35 | 6.77 | |
| BSAccess=Only DF2025.12 | 100 | 100 | 94.59 | 0.38 | 6.76 | |
| SalUnAccess=Only DF2025.12 | 100 | 100 | 94.57 | 0.33 | 6.77 | |
| FTBackbone=ResNet2026.04 | 98.2 | 98.9 | 93.3 | 71.6 | 0.029 | |
| SCRUBBackbone=ResNet2026.04 | 97.4 | 99.7 | 93.6 | 77 | 0.041 | |
| SALUNBackbone=ResNet2026.04 | 95.2 | 96.5 | 90.6 | 71.8 | 0.021 | |
| IEU w/GABackbone=ResNet2026.04 | 95.2 | 99.1 | 91.3 | 78.3 | 0.033 | |
| AmunAccess=Only DF2025.12 | 92.77 | 95.66 | 89.43 | 14.13 | 2.52 | |
| RetrainAccess=Only DF2025.12 | 92.09 | 100 | 91.85 | 16.78 | 0 | |
| Amun+SalunAccess=Only DF2025.12 | 91.9 | 96.59 | 89.98 | 13.07 | 2.35 | |
| IEU w/GA+NoisyBackbone=ResNet2026.04 | 91.5 | 98.2 | 90.4 | 71.9 | 0.007 | |
| RLBackbone=ResNet2026.04 | 90.7 | 97.3 | 88.7 | 71.4 | 0.011 | |
| IEU w/NoisyBackbone=ResNet2026.04 | 90.4 | 97.8 | 89.5 | 74.8 | 0.015 | |
| RetrainingBackbone=ResNet2026.04 | 90.2 | 99.3 | 90.3 | 71.4 | 0 |