Image Classification (Class Unlearning) on CIFAR-10
0Activation Distance (Dr)Finetune
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FinetuneBackbone=ResNet182024.11 | 0 | 1.66 | 0.03 | 1.46 | |
| BadTBackbone=ResNet182024.11 | 0 | 0.62 | 0.03 | 0.57 | |
| SCRUBBackbone=ResNet182024.11 | 0 | 0.75 | 0.04 | 0.69 | |
| Residual Feature AlignmentBackbone=ResNet182024.11 | 0 | 0.63 | 0.03 | 0.58 | |
| FinetuneBackbone=ViT2024.11 | 0 | 1.8 | 0.02 | 1.79 | |
| BadTBackbone=ViT2024.11 | 0 | 0.67 | 0.02 | 0.68 | |
| SCRUBBackbone=ViT2024.11 | 0 | 0.69 | 0.02 | 0.69 | |
| Residual Feature AlignmentBackbone=ViT2024.11 | 0 | 0.69 | 0.02 | 0.68 | |
| NegGradBackbone=ResNet182024.11 | 0.27 | 1.13 | 0.3 | 1.06 | |
| NegGradBackbone=ViT2024.11 | 1.11 | 0.94 | 1.13 | 0.96 |