Machine Unlearning on Tiny-ImageNet (Class-wise forgetting)
99.98Retention AccuracyRetrain
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Retrain2026.05 | 99.98 | 0 | 61.89 | 0 | 100 | 189.8 | |
| UAM2026.05 | 99.97 | 0.23 | 60.86 | 1.27 | 99.97 | 12.85 | |
| ROSU2026.05 | 99.97 | 0 | 61.89 | 0.01 | 100 | 14.29 | |
| PCGrad2026.05 | 99.84 | 0.08 | 59.59 | 2.52 | 100 | 9.14 | |
| OrthoGrad2026.05 | 99.67 | 0.13 | 58.96 | 3.37 | 100 | 17.64 | |
| GU2026.05 | 99.54 | 0 | 58.59 | 3.74 | 100 | 17.12 | |
| FT2026.05 | 98.54 | 26.88 | 57.52 | 32.69 | 97.37 | 9.41 | |
| l1-sparse2026.05 | 98.19 | 0 | 59.66 | 4.02 | 100 | 10.47 | |
| IU2026.05 | 95.83 | 1.2 | 55.03 | 12.21 | 99.84 | 1.27 | |
| NG2026.05 | 92.84 | 0.23 | 52.28 | 16.98 | 99.92 | 9.43 |