Machine Unlearning on ImageNet Retain set
54.67Zero-Shot Retention AccuracyOriginal
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OriginalBackbone=RN1012026.05 | 54.67 | — | — | |
| ICEDBackbone=RN1012026.05 | 53.67 | — | 98.17 | |
| CLIP-LIPBackbone=RN502026.05 | 52.77 | — | 100 | |
| CLIP-LIPBackbone=RN1012026.05 | 50.84 | — | 92.99 | |
| ICEDBackbone=RN502026.05 | 50.68 | — | 100 | |
| TIFSBackbone=RN1012026.05 | 50.17 | — | 91.77 | |
| OriginalBackbone=RN502026.05 | 48.67 | — | — | |
| TIFSBackbone=RN502026.05 | 44.67 | — | 91.78 | |
| GABackbone=RN1012026.05 | 38.76 | — | 70.9 | |
| EMMNBackbone=RN502026.05 | 36.67 | — | 75.34 | |
| GABackbone=RN502026.05 | 35.4 | — | 72.73 | |
| EMMNBackbone=RN1012026.05 | 29.33 | — | 53.65 | |
| LIPBackbone=RN502026.05 | 1.75 | — | 3.6 | |
| FTBackbone=RN1012026.05 | 0.51 | — | 0.93 | |
| FisherBackbone=RN1012026.05 | 0.38 | — | 0.7 | |
| FisherBackbone=RN502026.05 | 0.34 | — | 0.7 | |
| LIPBackbone=RN1012026.05 | 0.27 | — | 0.49 | |
| FTBackbone=RN502026.05 | 0.26 | — | 0.53 | |
| DUCKBackbone=ViT-b162025.03 | — | 99.39 | — | |
| FISHERBackbone=ViT-b162025.03 | — | 94.14 | — | |
| FTBackbone=ViT-b162025.03 | — | 99.92 | — | |
| GDR-GMABackbone=ViT-b162025.03 | — | 97.56 | — | |
| InfluenceBackbone=ViT-b162025.03 | — | 91.54 | — | |
| NegGradBackbone=ViT-b162025.03 | — | 89.63 | — | |
| NegGrad+Backbone=ViT-b162025.03 | — | 90.67 | — | |
| OriginalBackbone=ViT-b162025.03 | — | 94.04 | — | |
| OrthoGradBackbone=ViT-b162025.03 | — | 94.26 | — | |
| RetrainBackbone=ViT-b162025.03 | — | 95.08 | — | |
| SCRUBBackbone=ViT-b162025.03 | — | 96.18 | — |