Machine Unlearning on CINIC-10 CIFAR-10 Random Sampling Proxy-Retain (train test)
96.1Unlearning Accuracy (Au)Original
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OriginalArchitecture=ResNet182025.03 | 96.1 | 49.45 | 81.97 | — | |
| OrthoGradArchitecture=ResNet182025.03 | 80.99 | 61.69 | 68.41 | 8.9 | |
| GDR-GMAArchitecture=ResNet182025.03 | 79.93 | 93.97 | 67.02 | 10.4 | |
| FTArchitecture=ResNet182025.03 | 78.61 | 72.3 | 67.78 | — | |
| SCARArchitecture=ResNet182025.03 | 78.05 | 42.79 | 68.32 | 10.7 | |
| DUCKArchitecture=ResNet182025.03 | 53.22 | 99.47 | 46.43 | 39.2 | |
| NegGradArchitecture=ResNet182025.03 | 43.63 | 24.3 | 37.24 | 50.7 | |
| SCRUBArchitecture=ResNet182025.03 | 40.21 | 42.56 | 38.63 | 51.9 | |
| RetrainArchitecture=ResNet182025.03 | 29.95 | 99.7 | 30.48 | — | |
| NegGrad+Architecture=ResNet182025.03 | 21.43 | 28.36 | 19.89 | 74.8 | |
| FISHERArchitecture=ResNet182025.03 | 10.53 | 10.28 | 10.18 | 87.4 | |
| InfluenceArchitecture=ResNet182025.03 | 10.22 | 10.09 | 10 | 87.7 |