Machine Unlearning on Hubble Y YAGO
100FVM (ROUGE)Full (pre-unlearn)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Full (pre-unlearn)2026.05 | 100 | 50.1 | |
| GradDiff+RTRetain Training=true2026.05 | 98.8 | 50 | |
| NPO+RTRetain Training=true2026.05 | 96.4 | 50.3 | |
| RMU+RTRetain Training=true2026.05 | 85 | 50.9 | |
| SimNPO+RTRetain Training=true2026.05 | 83.5 | 49.9 | |
| CEU+RTRetain Training=true2026.05 | 43 | 50.1 | |
| GradAscent2026.05 | 39.4 | 50.3 | |
| Target (retrained)2026.05 | 11.9 | 51.5 | |
| SHRED2026.05 | 11.3 | 51.2 |