Machine Unlearning on CIFAR-100 Random Forget (10%)
99.98RASFRon
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SFRonBackbone=ResNet-18, Gamma=1/42026.01 | 99.98 | 62.69 | 66.7 | 38.31 | 1.94 | — | |
| FalWBackbone=ResNet-18, Gamma=1/42026.01 | 99.98 | 65.12 | 65.86 | 33.58 | 0.89 | — | |
| RetrainBackbone=ResNet-18, Gamma=1/42026.01 | 99.95 | 64.82 | 64.24 | 35.18 | 0 | — | |
| SalUnBackbone=ResNet-18, Gamma=1/42026.01 | 99.95 | 65.8 | 63.39 | 32.2 | 1.2 | — | |
| Retrain2025.09 | 99.95 | — | 61.24 | 58.35 | 0 | 38.12 | |
| GD+SineParams (%)=0.92025.09 | 99.93 | — | 60.85 | 59.14 | 2.35 | 37.48 | |
| FTBackbone=ResNet-18, Gamma=1/42026.01 | 99.92 | 96.27 | 70.3 | 3.73 | 17.24 | — | |
| GD+TanhParams (%)=0.92025.09 | 99.9 | — | 60.4 | 59.6 | 2.5 | 37.9 | |
| CoUnParams (%)=100.02025.09 | 99.85 | — | 60.52 | 59.42 | 2.58 | 38.24 | |
| SalUnParams (%)=100.02025.09 | 99.21 | — | 59.42 | 61.28 | 3.48 | 39.85 | |
| FTParams (%)=100.02025.09 | 98.42 | — | 57.14 | 65.42 | 5.82 | 41.25 | |
| GD+LoRAParams (%)=0.92025.09 | 97.85 | — | 53.82 | 78.21 | 10.05 | 52.31 | |
| RLBackbone=ResNet-18, Gamma=1/42026.01 | 97.52 | 31.07 | 62.22 | 48.93 | 12.99 | — | |
| IUBackbone=ResNet-18, Gamma=1/42026.01 | 96.93 | 96.87 | 70.49 | 3.13 | 18.34 | — | |
| BEBackbone=ResNet-18, Gamma=1/42026.01 | 96.81 | 96.71 | 69.61 | 3.29 | 18.07 | — | |
| GABackbone=ResNet-18, Gamma=1/42026.01 | 96.66 | 95.22 | 68.83 | 4.78 | 17.17 | — | |
| BSBackbone=ResNet-18, Gamma=1/42026.01 | 96.6 | 96.53 | 68.88 | 3.47 | 17.85 | — | |
| GAParams (%)=100.02025.09 | 95.18 | — | 51.35 | 72.85 | 12.41 | 48.72 | |
| l1-sparseBackbone=ResNet-18, Gamma=1/42026.01 | 91.38 | 91.24 | 64.14 | 8.76 | 15.38 | — |