Machine Unlearning on CIFAR-100 (50% random data removal)
99.98Retain AccuracyRetrain
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RetrainModel=ResNet-182025.09 | 99.98 | 31.41 | 68.41 | 58.35 | 0 | 15.24 | |
| FTModel=ResNet-182025.09 | 99.98 | 17.36 | 74.16 | 50.6 | 6.89 | 4.19 | |
| NegGrad+Model=ResNet-182025.09 | 99.98 | 26.32 | 71.98 | 52.32 | 3.67 | 5.36 | |
| CoUnModel=ResNet-182025.09 | 99.98 | 31.43 | 65.6 | 55.99 | 1.3 | 5.58 | |
| RetrainModel=ViT2025.09 | 99.98 | 48.07 | 52.4 | 69.54 | 0 | 48.35 | |
| ℓ1-sparseModel=ResNet-182025.09 | 99.94 | 32.26 | 67.66 | 51.54 | 2.11 | 4.19 | |
| CoUnModel=VGG-162025.09 | 99.88 | 42.37 | 55.19 | 50 | 0.85 | 3.18 | |
| CoUnModel=ViT2025.09 | 99.71 | 45.95 | 49.45 | 59.24 | 3.91 | 10.74 | |
| RetrainModel=VGG-162025.09 | 99.65 | 42.85 | 57.7 | 50.19 | 0 | 8.67 | |
| NegGrad+Model=ViT2025.09 | 99.3 | 45.35 | 50.82 | 55.07 | 4.86 | 6.45 | |
| SalUnModel=ViT2025.09 | 98.93 | 45.64 | 39.46 | 76.49 | 5.84 | 12.03 | |
| FTModel=ViT2025.09 | 98.71 | 10.91 | 56.79 | 28.18 | 21.05 | 1.61 | |
| NoTModel=ResNet-182025.09 | 98.64 | 26.43 | 67.97 | 43.82 | 5.32 | 2.01 | |
| ℓ1-sparseModel=VGG-162025.09 | 98.25 | 34.24 | 62.76 | 42.12 | 5.79 | 1.91 | |
| NoTModel=ViT2025.09 | 97.86 | 31.81 | 55.51 | 48.85 | 10.55 | 3.22 | |
| FTModel=VGG-162025.09 | 97.71 | 29.82 | 63.72 | 39.98 | 7.8 | 1.43 | |
| SalUnModel=ResNet-182025.09 | 95.61 | 25.43 | 60.35 | 57.14 | 4.91 | 1.95 | |
| NegGrad+Model=VGG-162025.09 | 95.54 | 43.42 | 58.52 | 43.51 | 3.04 | 3.18 | |
| NoTModel=VGG-162025.09 | 94.23 | 34.64 | 61.58 | 39.84 | 6.96 | 2.38 | |
| SalUnModel=VGG-162025.09 | 91.98 | 37.6 | 57.3 | 53.84 | 4.24 | 2.45 | |
| ℓ1-sparseModel=ViT2025.09 | 71.18 | 47.3 | 53.32 | 44.22 | 13.95 | 8.06 |