Machine Unlearning on Tiny-ImageNet (10% Random Forgetting)
51.06Unlearning Accuracy (UA)NegGrad+
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| NegGrad+Backbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 51.06 | 83.22 | 46.74 | 51.97 | 8.65 | 6.58 | — | |
| MCU-βBackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 49.92 | 92.88 | 52.92 | 46.9 | 5.67 | 10.83 | — | |
| RTBackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 45.45 | 99.52 | 55.59 | 55.79 | 0 | 37.46 | — | |
| MCUBackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 42.42 | 93.32 | 52.53 | 44.43 | 5.91 | 10.77 | — | |
| RLBackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 38.59 | 99.03 | 53.87 | 86.53 | 9.95 | 13.33 | — | |
| NegGrad+Backbone=ResNet-182025.09 | 38.01 | 99.98 | 64.68 | 57.84 | 2.15 | — | 80.21 | |
| l1-sparseBackbone=ResNet-182025.09 | 36.96 | 99.96 | 62.62 | 56.74 | 2.25 | — | 60.16 | |
| SalUnBackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 36.61 | 99.03 | 54.04 | 85.37 | 10.12 | 13.79 | — | |
| RetrainBackbone=ResNet-182025.09 | 36.16 | 99.98 | 63.82 | 63.73 | 0 | — | 218.98 | |
| NoTBackbone=ResNet-182025.09 | 35.64 | 99.98 | 63.66 | 56.08 | 2.08 | — | 80.21 | |
| CoUnBackbone=ResNet-182025.09 | 35.1 | 99.95 | 63.27 | 57.57 | 1.95 | — | 80.21 | |
| SalUnBackbone=ResNet-182025.09 | 34.03 | 98.52 | 61.21 | 67.72 | 2.55 | — | 51.08 | |
| FTBackbone=ResNet-182025.09 | 32.76 | 99.98 | 64.65 | 56.93 | 2.76 | — | 60.16 | |
| SFRonBackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 28.87 | 97.14 | 53.1 | 15.29 | 15.49 | 12.6 | — | |
| FTBackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 5.76 | 99.34 | 56.25 | 15.95 | 20.09 | 3.8 | — | |
| GABackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 5.17 | 96.11 | 53.66 | 7.89 | 23.38 | 0.32 | — | |
| NegTVBackbone=VGG-16-BN, Forgetting Ratio=10%2025.05 | 0.81 | 99.35 | 56.85 | 4.49 | 24.34 | 0.58 | — |