Machine Unlearning on AGNews Dt (test)
90.44AccuracyFisher Forgetting
Evaluation Results
| Method | Links | |
|---|---|---|
| Fisher Forgettingthreshold=0.352026.06 | 90.44 | |
| Certified Hessian Forgettingthreshold=0.32026.06 | 90.16 | |
| Adversarial Retrainthreshold=0.32026.06 | 89.98 | |
| Pre-unlearn2026.06 | 89.9 | |
| Fine-tunethreshold=0.22026.06 | 89.85 | |
| Relabel with Fine-tunethreshold=0.32026.06 | 89.74 | |
| Retrainthreshold=0.32026.06 | 89.6 | |
| Gradient Ascent with Fine-tunethreshold=0.32026.06 | 89.55 | |
| Hessian Forgettingthreshold=0.32026.06 | 82.7 | |
| Gradient Ascentthreshold=0.052026.06 | 64.28 | |
| Relabelthreshold=0.052026.06 | 58.54 |