Classification Unlearning on CIFAR-100 (10% Random Data Forgetting)
55.03Utility Accuracy (UA)RL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RLBackbone=ResNet-18, TParams=100%2026.05 | 55.03 | 99.81 | 70.03 | 98.97 | |
| SalUnBackbone=ResNet-18, TParams=50%2026.05 | 27.53 | 97 | 67.79 | 70.79 | |
| RetrainBackbone=ResNet-18, TParams=100%2026.05 | 26.47 | 99.97 | 74.13 | 51 | |
| SalUn-softBackbone=ResNet-18, TParams=50%2026.05 | 24.24 | 98.95 | 70.48 | 79.13 | |
| l1-sparseBackbone=ResNet-18, TParams=100%2026.05 | 10.64 | 96.62 | 70.99 | 22.58 | |
| IUBackbone=ResNet-18, TParams=100%2026.05 | 3.18 | 97.15 | 73.49 | 9.62 | |
| GABackbone=ResNet-18, TParams=100%2026.05 | 3.13 | 97.33 | 75.31 | 7.24 | |
| SEMU-remainBackbone=ResNet-18, TParams=1.18%2026.05 | 2.93 | 97.33 | 74.16 | 11.93 | |
| BARRIERBackbone=ResNet-18, TParams=0.46%2026.05 | 2.8 | 97.51 | 76.18 | 5.96 | |
| SEMUBackbone=ResNet-18, TParams=1.18%2026.05 | 2.53 | 97.39 | 74.14 | 8.82 | |
| FTBackbone=ResNet-18, TParams=100%2026.05 | 2.42 | 99.95 | 75.55 | 11.04 | |
| BEBackbone=ResNet-18, TParams=100%2026.05 | 2.31 | 97.27 | 73.93 | 9.62 | |
| BSBackbone=ResNet-18, TParams=100%2026.05 | 2.27 | 97.41 | 75.26 | 5.82 |