Image Classification Unlearning on CIFAR-100 30% random data forgetting
99.6Utility Retention (Retained Train)SCRUB
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SCRUBBackbone=ResNet2026.04 | 99.6 | 89.2 | 72.7 | 81.1 | 7.6 | |
| IEU w/GABackbone=ResNet2026.04 | 98.7 | 77.7 | 68 | 77.8 | 2.6 | |
| RetrainingBackbone=ResNet2026.04 | 98.5 | 67.7 | 68 | 78.1 | 0 | |
| RLBackbone=ResNet2026.04 | 98.5 | 68.8 | 67.7 | 77.9 | 0.4 | |
| FTBackbone=ResNet2026.04 | 98.4 | 83.4 | 69.9 | 77.8 | 4.5 | |
| IEU w/GA+NoisyBackbone=ResNet2026.04 | 96.6 | 68.4 | 66.6 | 77.4 | 1.2 | |
| IEU w/NoisyBackbone=ResNet2026.04 | 96 | 67.9 | 67.1 | 77.7 | 1 | |
| SALUNBackbone=ResNet2026.04 | 75.8 | 57.3 | 55.5 | 82.7 | 12.6 |