Machine Unlearning on Hubble Gutenberg
0.226FVM (ROUGE)SimNPO+RT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SimNPO+RTRetain Training=true2026.05 | 0.226 | 0.496 | |
| Full (pre-unlearn)2026.05 | 0.197 | 0.501 | |
| RMU+RTRetain Training=true2026.05 | 0.192 | 0.502 | |
| NPO+RTRetain Training=true2026.05 | 0.184 | 0.502 | |
| GradAscent2026.05 | 0.176 | 0.505 | |
| SHRED2026.05 | 0.176 | 0.505 | |
| CEU+RTRetain Training=true2026.05 | 0.175 | 0.502 | |
| GradDiff+RTRetain Training=true2026.05 | 0.174 | 0.502 | |
| Target (retrained)2026.05 | 0.169 | 0.515 |