Machine Unlearning on PathMNIST Internal (test)
97.78Accuracy (Retained)Finetune
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FinetuneBackbone=ViT, Data requirement=Requires retain set2025.11 | 97.78 | 0 | 1.1 | 5 | 46.67 | |
| FCSBackbone=ViT, Data requirement=Requires retain set2025.11 | 88.81 | 0 | 1.01 | 13 | 49 | |
| Retrained ModelBackbone=ViT, Data requirement=Retraining (classifier head only)2025.11 | 88.7 | 0 | 1.01 | — | — | |
| Original ModelBackbone=ViT, Data requirement=Full Dataset2025.11 | 87.49 | 96.83 | 0.51 | — | 50.43 | |
| POUR-PBackbone=ViT, Data requirement=Forget set only2025.11 | 87.14 | 0 | 1 | — | 50.43 | |
| Gradient AscentBackbone=ViT, Data requirement=Forget set only2025.11 | 81.43 | 9.03 | 0.86 | 22 | 48.71 | |
| POUR-DBackbone=ViT, Data requirement=Forget set only2025.11 | 81.09 | 7.88 | 0.87 | 63 | 51 | |
| Random LabelBackbone=ViT, Data requirement=Forget set only2025.11 | 78.71 | 23.73 | 0.74 | 26 | 48.23 | |
| DELETEBackbone=ViT, Data requirement=Forget set only2025.11 | 72.75 | 0 | 0.85 | 50 | 49.76 |