Subclass-level unlearning on CIFAR-100 (test)
90Accuracy (Df)Original
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OriginalBackbone=ResNet-182026.03 | 90 | 80 | 85.33 | |
| FTBackbone=ResNet-182026.03 | 62.33 | 77.83 | 83.89 | |
| GABackbone=ResNet-182026.03 | 56 | 59 | 80.57 | |
| ℓ1-sparseBackbone=ResNet-182026.03 | 51.67 | 82.42 | 84.64 | |
| SSDBackbone=ResNet-182026.03 | 33 | 39.25 | 67.23 | |
| MunbaBackbone=ResNet-182026.03 | 31.67 | 69.75 | 75.32 | |
| GDRBackbone=ResNet-182026.03 | 8.67 | 22.33 | 79.93 | |
| SCRUBBackbone=ResNet-182026.03 | 7 | 54.75 | 75.42 | |
| SalUnBackbone=ResNet-182026.03 | 3 | 34.9 | 71.78 | |
| Our methodBackbone=ResNet-182026.03 | 2.33 | 78.17 | 81.1 | |
| DELETEBackbone=ResNet-182026.03 | 0.67 | 2.83 | 82.09 |