Multi-class unlearning on CIFAR-100 5 Classes
77.01Accuracy (Fine-tuned)Gradient Ascent
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Gradient AscentBackbone=ViT-Small, Rank=322026.06 | 77.01 | 83.31 | 94.7 | |
| L2ULBackbone=ViT-Small, Rank=322026.06 | 75.86 | 83.26 | 94 | |
| SalUnBackbone=ViT-Small, Rank=322026.06 | 71.84 | 97.17 | 12.4 | |
| FinetuneBackbone=ViT-Small, Rank=322026.06 | 52.82 | 86.52 | 72.4 | |
| Boundary ExpandBackbone=ViT-Small, Rank=322026.06 | 51.47 | 94.27 | 1.2 | |
| Boundary ShrinkBackbone=ViT-Small, Rank=322026.06 | 46.81 | 82.44 | 68.7 | |
| Random LabelBackbone=ViT-Small, Rank=322026.06 | 7.67 | 75.88 | 6.1 | |
| DeleteBackbone=ViT-Small, Rank=322026.06 | 2.68 | 97.1 | 9.4 | |
| Bad TeacherBackbone=ViT-Small, Rank=322026.06 | 2.24 | 79.01 | 100 | |
| TrustEraseBackbone=ViT-Small, Rank=322026.06 | 0 | 88.7 | 0 |