Machine Unlearning Evaluation on TOFU (Real Authors)
93.3ROUGEWISE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| WISECategory=Model Editing, Target=Dummy2025.12 | 93.3 | 43.7 | 58.7 | |
| WISECategory=Model Editing, Target=Incorrect2025.12 | 93.3 | 42.3 | 55.4 | |
| WISECategory=Model Editing, Target=Avoidant2025.12 | 93.3 | 40.6 | 56.4 | |
| Ground TruthDescription=Retain 90 Llama-2-7B model2025.12 | 92.3 | 43.4 | 57.1 | |
| GDCategory=Model Unlearning2025.12 | 81.7 | 56.4 | 73.2 | |
| POCategory=Model Unlearning2025.12 | 52.5 | 40.3 | 0 | |
| IKECategory=Model Editing, Target=Incorrect2025.12 | 48 | 35.5 | 41.7 | |
| IKECategory=Model Editing, Target=Avoidant2025.12 | 47.4 | 33 | 36.8 | |
| IKECategory=Model Editing, Target=Dummy2025.12 | 16.3 | 34.1 | 38.9 | |
| ROMECategory=Model Editing, Target=Dummy2025.12 | 12.6 | 35.4 | 45.4 | |
| ROMECategory=Model Editing, Target=Avoidant2025.12 | 0.7 | 27.5 | 28.1 | |
| ROMECategory=Model Editing, Target=Incorrect2025.12 | 0.2 | — | — | |
| KLCategory=Model Unlearning2025.12 | 0 | 27.4 | 39.2 | |
| GACategory=Model Unlearning2025.12 | 0 | 23.2 | 36.8 |