Machine Unlearning on CIFAR-10 10% random data forgetting
56.12Utility Accuracy (UA)l1-sparse
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| l1-sparseBackbone=ResNet-182026.05 | 56.12 | 44.04 | 41.98 | 21.43 | 44.15 | |
| RLBackbone=ResNet-182026.05 | 18.75 | 82.38 | 78.75 | 32.28 | 18.3 | |
| GABackbone=ResNet-182026.05 | 11.29 | 88.73 | 84.02 | 1.93 | 16.9 | |
| SalUnBackbone=ResNet-182026.05 | 7.68 | 93.07 | 88.23 | 14.75 | 8.72 | |
| HeldDist-RAUL-LSBackbone=ResNet-18, Gradient Aggregator=LS2026.05 | 7.3 | 99.24 | 92.09 | 17.21 | 2.72 | |
| HeldDist-RAUL-UPGradBackbone=ResNet-18, Gradient Aggregator=UPGrad2026.05 | 6.47 | 99.38 | 92.38 | 15.31 | 2.23 | |
| RetrainBackbone=ResNet-182026.05 | 5.68 | 100 | 94.32 | 13.13 | 0 | |
| FTBackbone=ResNet-182026.05 | 3.93 | 97.56 | 92.22 | 8.15 | 8.44 | |
| UniDist-RAUL-UPGradBackbone=ResNet-18, Gradient Aggregator=UPGrad2026.05 | 1.43 | 99.11 | 91.95 | 9.52 | 2.53 | |
| UniDist-RAUL-LSBackbone=ResNet-18, Gradient Aggregator=LS2026.05 | 0.97 | 99.2 | 91.56 | 7.44 | 3.24 | |
| IUBackbone=ResNet-182026.05 | 0.74 | 99.15 | 93.32 | 1.95 | 6.06 |