Machine Unlearning on AGNews (remaining set Dr)
99.95AccuracyAdversarial Retrain
Evaluation Results
| Method | Links | |
|---|---|---|
| Adversarial Retrainthreshold=0.32026.06 | 99.95 | |
| Fine-tunethreshold=0.22026.06 | 99.95 | |
| Relabel with Fine-tunethreshold=0.32026.06 | 99.95 | |
| Gradient Ascent with Fine-tunethreshold=0.32026.06 | 99.95 | |
| Retrainthreshold=0.32026.06 | 99.92 | |
| Certified Hessian Forgettingthreshold=0.32026.06 | 99.89 | |
| Pre-unlearn2026.06 | 99.85 | |
| Fisher Forgettingthreshold=0.352026.06 | 99.78 | |
| Hessian Forgettingthreshold=0.32026.06 | 88.91 | |
| Gradient Ascentthreshold=0.052026.06 | 68.24 | |
| Relabelthreshold=0.052026.06 | 67.09 |