Machine Unlearning Effectiveness Assessment on CIFAR-10 (UA/MIA Focus)
100Unlearning Accuracy (UA)RT
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| RTBackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 100 | 100 | 99.98 | 90.37 | 100 | 0 | 104.93 | |
| RLBackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 100 | 100 | 96.67 | 87.81 | 100 | 1.17 | 6.8 | |
| SFRonBackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 100 | 100 | 83.46 | 81.8 | 100 | 5.02 | 18.59 | |
| SalUnBackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 100 | 100 | 99.81 | 90.34 | 100 | 0.04 | 6.97 | |
| MCUβBackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 100 | 100 | 99.85 | 90.37 | 100 | 0.03 | 7.29 | |
| MCUBackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 99.96 | 100 | 99.8 | 90.37 | 100 | 0.04 | 7.27 | |
| NegGrad+Backbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 99.94 | 100 | 98.07 | 87.25 | 99.97 | 1.02 | 2.95 | |
| GABackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 85.01 | 88.5 | 90.55 | 80.9 | 86.27 | 11.82 | 0.4 | |
| NegTVBackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 25.28 | 31.95 | 93.03 | 82.43 | 29.05 | 45.72 | 0.71 | |
| FTBackbone=PreResNet-110, Unlearning Setting=Class-wise Forgetting2025.05 | 18.53 | 24.63 | 99.94 | 91.01 | 43.18 | 42.87 | 5.52 |