Class Unlearning on CIFAR-10 (ToW, DA, NMI Evaluation)
0.999ToWRETRAIN
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RETRAINBackbone=ResNet182026.06 | 0.999 | 0.987 | 0.41 | |
| ENGRAMBackbone=ResNet18, alpha=best2026.06 | 0.984 | 0.958 | 0.611 | |
| l1-SPARSEBackbone=ResNet182026.06 | 0.956 | 0.975 | 0.515 | |
| FINE-TUNEBackbone=ResNet182026.06 | 0.952 | 0.973 | 0.547 | |
| NEGGRAD+Backbone=ResNet182026.06 | 0.936 | 0.942 | 0.244 | |
| ENGRAMBackbone=ResNet18, alpha=12026.06 | 0.93 | 0.992 | 0.379 | |
| RANDOM LABELBackbone=ResNet182026.06 | 0.916 | 0.851 | 0.996 | |
| SALUNBackbone=ResNet182026.06 | 0.878 | 0.833 | 0.911 | |
| NEGGRADBackbone=ResNet182026.06 | 0.711 | 0.97 | 0.037 |