Machine Unlearning on CIFAR10 unlearning (Du)
99.93AccuracyCert.-Hess.
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Cert.-Hess.Proportion of remaining data used=0.32026.06 | 99.93 | 67.25 | 1.91 | |
| Pre-unlearn2026.06 | 95.92 | — | — | |
| FisherProportion of remaining data used=0.32026.06 | 95.87 | 68.96 | 1.37 | |
| HessianProportion of remaining data used=0.352026.06 | 92.67 | 68.58 | 1.37 | |
| Fine-tuneProportion of remaining data used=0.32026.06 | 76.13 | 70.53 | 1.42 | |
| Adv. Retr.Proportion of remaining data used=0.352026.06 | 73.87 | 71.1 | 1.22 | |
| Relab.+FTProportion of remaining data used=0.32026.06 | 73.67 | 70.94 | 1.3 | |
| GA+FTProportion of remaining data used=0.32026.06 | 73.4 | 70.33 | 1.36 | |
| RetrainProportion of remaining data used=0.352026.06 | 69.15 | 71.23 | 1.22 | |
| RelabelProportion of remaining data used=0.052026.06 | 46.88 | 30.9 | 7.77 | |
| GAProportion of remaining data used=0.052026.06 | 44.26 | 35.17 | 6.2 |