Machine Unlearning Utility and Risk Metrics on CIFAR-100
99.98Retention Accuracy (RA)Retrained (Baseline)
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Retrained (Baseline)Backbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 99.98 | 94.4 | 0.23 | 50 | 50 | — | — | — | |
| FTBackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 99.96 | 24.94 | 1.02 | 13 | 13 | 0.27 | 0.05 | 0.48 | |
| RELOADBackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 99.56 | 1.47 | 0.08 | 1 | 0 | 0.32 | 0.12 | 0.53 | |
| CF-kBackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 98.85 | 21.38 | 0.92 | 12 | 11 | 0.1 | 0.06 | 0.55 | |
| EU-kBackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 98.3 | 20.18 | 0.89 | 11 | 11 | 0.21 | 0.07 | 0.56 | |
| FisherBackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 97.76 | 1.54 | 0.08 | 3 | 3 | 1.77 | 0.07 | 0.56 | |
| SalUnBackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 97.33 | 40.31 | 1.2 | 10 | 10 | 0.2 | 0.06 | 0.55 | |
| GABackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 36.61 | 45.98 | 4.71 | 6 | 6 | 0.01 | 0.06 | 0.66 | |
| SCRUBBackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 23.96 | 48.86 | 1.97 | 1 | 1 | 0.07 | 0.06 | 0.65 | |
| SSDBackbone=ResNet-18, Forget Rate=30%, Forgetting Type=Random Forgetting2026.04 | 11.89 | 65.53 | 3.15 | 3 | 2 | 0.01 | 8.3 | 7.83 |