Machine Unlearning on CIFAR-10 Class-wise forgetting
99.99Retention Accuracy (RA)Retrain
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Retrain2026.05 | 99.99 | 0 | 95.15 | 0 | 100 | 33.8 | |
| ROSU2026.05 | 99.96 | 0 | 95.01 | 0.17 | 100 | 2.98 | |
| UAM2026.05 | 99.91 | 0.02 | 94.42 | 0.83 | 100 | 2.47 | |
| FF2026.05 | 98.95 | 10.98 | 93.26 | 13.91 | 100 | 1.07 | |
| OrthoGrad2026.05 | 98.59 | 0.46 | 92.31 | 4.7 | 99.66 | 7.62 | |
| PCGrad2026.05 | 98.58 | 0.07 | 92.93 | 3.7 | 100 | 2.1 | |
| FT2026.05 | 98.22 | 24.94 | 92.68 | 29.18 | 99.13 | 1.63 | |
| ℓ1-sparse2026.05 | 97.18 | 0 | 92.07 | 5.89 | 100 | 1.73 | |
| GU2026.05 | 94.2 | 6.48 | 88.72 | 18.7 | 95.19 | 7.59 | |
| IU2026.05 | 90.86 | 10.28 | 85.86 | 28.7 | 98.27 | 0.59 | |
| NG2026.05 | 88.03 | 0.84 | 83 | 24.95 | 99.28 | 1.64 |