Image Classification (Sample Unlearning) on Fashion-MNIST
0Activation Distance (Dr)NegGrad
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| NegGradBackbone=ResNet182024.11 | 0 | 0.07 | 0.03 | — | |
| FinetuneBackbone=ResNet182024.11 | 0 | 0.04 | 0.03 | — | |
| BadTBackbone=ResNet182024.11 | 0 | 0.78 | 0.03 | — | |
| SCRUBBackbone=ResNet182024.11 | 0 | 0.05 | 0.03 | — | |
| Residual Feature AlignmentBackbone=ResNet182024.11 | 0 | 0.07 | 0.03 | — | |
| FinetuneBackbone=ViT2024.11 | 0 | 0.05 | 0.03 | — | |
| SCRUBBackbone=ViT2024.11 | 0 | 0.06 | 0.03 | — | |
| Residual Feature AlignmentBackbone=ViT2024.11 | 0 | 0.09 | 0.03 | — | |
| NegGradBackbone=ViT2024.11 | 0.01 | 0.06 | 0.03 | — | |
| BadTBackbone=ViT2024.11 | 0.01 | 0.05 | 0.03 | — | |
| NegGradBackbone=ResNet182024.11 | 0.42 | 0.39 | 0.43 | — | |
| SCRUBBackbone=ResNet182024.11 | 0.66 | 0.82 | 0.65 | — | |
| BadTBackbone=ResNet182024.11 | 0.83 | 0.82 | 0.8 | — | |
| FinetuneBackbone=ResNet182024.11 | 0.85 | 0.84 | 0.84 | — | |
| Residual Feature AlignmentBackbone=ResNet182024.11 | 1.11 | 1.25 | 1.13 | — | |
| RetrainBackbone=ResNet182024.11 | 2.3 | 22.7 | 2.28 | — | |
| BadTBackbone=ResNet182024.11 | — | — | — | 5 | |
| BadTBackbone=ViT2024.11 | — | — | — | 100 | |
| FinetuneBackbone=ResNet182024.11 | — | — | — | 100 | |
| FinetuneBackbone=ViT2024.11 | — | — | — | 100 | |
| NegGradBackbone=ResNet182024.11 | — | — | — | 100 | |
| NegGradBackbone=ViT2024.11 | — | — | — | 100 | |
| OriginalBackbone=ResNet182024.11 | — | — | — | 100 | |
| OriginalBackbone=ViT2024.11 | — | — | — | 100 | |
| Residual Feature Alignment UnlearningBackbone=ResNet182024.11 | — | — | — | 93 | |
| Residual Feature Alignment UnlearningBackbone=ViT2024.11 | — | — | — | 98 | |
| RetrainBackbone=ResNet182024.11 | — | — | — | 95 | |
| RetrainBackbone=ViT2024.11 | — | — | — | 100 | |
| SCRUBBackbone=ResNet182024.11 | — | — | — | 100 | |
| SCRUBBackbone=ViT2024.11 | — | — | — | 100 |