Machine Unlearning on CIFAR-100 50% class-wise data forgetting (train test)
99.9Accuracy Dr (Train)Retraining
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RetrainingBackbone=ViT2026.04 | 99.9 | 68.5 | 0 | 0 | 83.4 | 0 | |
| IEU w/GABackbone=ViT2026.04 | 99.9 | 68.8 | 0 | 0 | 83.4 | 0.001 | |
| FTBackbone=ViT2026.04 | 99.8 | 67.7 | 5.8 | 3.8 | 83.4 | 0.021 | |
| RLBackbone=ViT2026.04 | 99.6 | 66.5 | 0 | 0 | 83.3 | 0.005 | |
| SCRUBBackbone=ResNet2026.04 | 99.5 | 80.6 | 0 | 0 | 83.3 | 0.009 | |
| IEU w/GABackbone=ResNet2026.04 | 99.2 | 77.4 | 0 | 0 | 83.3 | 0.002 | |
| RLBackbone=ResNet2026.04 | 99.1 | 78.6 | 0 | 0 | 83.3 | 0.004 | |
| RetrainingBackbone=ResNet2026.04 | 98.9 | 77.3 | 0 | 0 | 83.7 | 0 | |
| FTBackbone=ResNet2026.04 | 98.9 | 79.7 | 0 | 0 | 83.3 | 0.006 | |
| IEU w/GA+NoisyBackbone=ResNet2026.04 | 98.3 | 77 | 0 | 0 | 83.3 | 0.003 | |
| IEU w/NoisyBackbone=ResNet2026.04 | 98 | 76.9 | 0 | 0 | 83.3 | 0.003 | |
| IEU w/GA+NoisyBackbone=ViT2026.04 | 98 | 67.5 | 0 | 0 | 83.6 | 0.006 | |
| IEU w/NoisyBackbone=ViT2026.04 | 97.7 | 67.9 | 0 | 0 | 83.5 | 0.006 | |
| SCRUBBackbone=ViT2026.04 | 84.2 | 67 | 0 | 0 | 83.6 | 0.035 | |
| SALUNBackbone=ResNet2026.04 | 70.7 | 60.9 | 1.2 | 1 | 83.3 | 0.094 | |
| SALUNBackbone=ViT2026.04 | 66.4 | 58.7 | 0.1 | 0.1 | 85.3 | 0.091 |