Machine Unlearning on ImageNet Random Sampling 3-seed average (train)
94.54Au ScoreSCRUB
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SCRUBArchitecture=ViT2025.03 | 94.54 | 96 | 80.76 | 8.4 | |
| FISHERArchitecture=ViT2025.03 | 94.36 | 94.01 | 81 | 8.2 | |
| OriginalArchitecture=ViT2025.03 | 94.26 | 94.04 | 81.06 | — | |
| SSDArchitecture=ViT2025.03 | 93.5 | 93.38 | 80.35 | 8.1 | |
| FTArchitecture=ViT2025.03 | 89.96 | 99.94 | 75.03 | — | |
| InfluenceArchitecture=ViT2025.03 | 89.08 | 91.61 | 77.48 | 7.1 | |
| OrthoGradArchitecture=ViT2025.03 | 81.04 | 97.59 | 78.22 | 1.8 | |
| GDR-GMAArchitecture=ViT2025.03 | 80.41 | 99.34 | 75.47 | 3.8 | |
| NegGrad+Architecture=ViT2025.03 | 78.75 | 87.63 | 72.16 | 6.9 | |
| DUCKArchitecture=ViT2025.03 | 76.75 | 100 | 69.45 | 9.7 | |
| SCARArchitecture=ViT2025.03 | 71.49 | 93.53 | 77.83 | 7.9 | |
| NegGradArchitecture=ViT2025.03 | 24.2 | 24.26 | 21.06 | 72 | |
| RetrainArchitecture=ViT2025.03 | 3.36 | 98.72 | 3.42 | — |