Machine Unlearning on TinyImageNet class-wise forgetting (test)
100Unlearning Accuracy (UA)RT
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| RTBackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 100 | 100 | 99.34 | 56.94 | 100 | 0 | 42.06 | |
| SFRonBackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 100 | 100 | 88.62 | 51.9 | 100 | 3.15 | 12.03 | |
| MCUBackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 100 | 100 | 99.1 | 56.44 | 100 | 0.15 | 5.78 | |
| MCUβBackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 100 | 100 | 99.07 | 56.47 | 100 | 0.15 | 5.77 | |
| SalUnBackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 99.27 | 100 | 98.95 | 56.58 | 100 | 0.3 | 7.25 | |
| RLBackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 98.87 | 100 | 98.83 | 56.52 | 100 | 0.41 | 7.24 | |
| NegGrad+Backbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 94.76 | 93.6 | 99.33 | 56.73 | 97.33 | 2.91 | 2.25 | |
| GABackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 91.8 | 87.33 | 94.75 | 52.86 | 96.6 | 6.59 | 0.13 | |
| FTBackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 74.27 | 78.67 | 99.29 | 56.71 | 90.53 | 11.36 | 4.29 | |
| NegTVBackbone=VGG-16-BN, Forgetting scenario=Class-wise Forgetting2025.05 | 0.5 | 50 | 99.38 | 56.96 | 6.1 | 48.69 | 0.2 |