Machine Unlearning on Real-world unlearning scenario
82.17Forget AccuracyOriginal
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OriginalModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 82.17 | 87.13 | 77.78 | |
| GAModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 39.53 | — | 58.51 | |
| NPOModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 39.53 | — | 47.41 | |
| CiPOModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 34.11 | 47.5 | 74.07 | |
| GDModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 29.45 | — | 28.14 | |
| AnswerIDKModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 26.36 | 84.81 | 29.63 | |
| R²MUModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 24.8 | — | 25.93 | |
| DirectIDKModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 3.88 | 6.36 | 5.93 | |
| ReasonedIDKModel=DeepSeek-R1-0528-Qwen3-8B2026.04 | 0.7 | 41.93 | 0 |