Machine Unlearning on Years
75.4MMLUMLP Breaking + CIR
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MLP Breaking + CIRModel=Qwen3-8B2026.05 | 75.4 | — | 39.4 | 40.6 | 1.1 | |
| MLP Breaking + CIR + MCUModel=Qwen3-8B2026.05 | 75.3 | — | 30.1 | 30.9 | 0.8 | |
| MLP Breaking + CIRModel=Gemma2-9B2026.05 | 69.6 | — | 30 | 36.3 | 6.3 | |
| MLP Breaking + CIR + MCUModel=Gemma2-9B2026.05 | 68.9 | — | 26.2 | 28.9 | 2.7 | |
| Original model2026.05 | 65.1 | 1 | 68.4 | — | — | |
| MLP BreakingCIR=true2026.05 | 64.6 | 1.01 | 32.7 | 39.4 | 6.7 | |
| MLP BreakingCIR=true, MCU=true2026.05 | 63.8 | 1.01 | 25.9 | 30.9 | 6.5 | |
| MLP Breaking2026.05 | 60.8 | 1.239 | 27 | 51.5 | 24.5 | |
| NPO2026.05 | 58.5 | 1.404 | 27.9 | 63.4 | 35.6 | |
| MLP BreakingSAM=true2026.05 | 58 | 1.215 | 25.6 | 47.4 | 21.8 | |
| RMUCIR=true, MCU=true2026.05 | 57.2 | 1.101 | 31.7 | 33.7 | 2 | |
| RMUCIR=true2026.05 | 57.1 | 1.102 | 30.7 | 37.4 | 6.7 | |
| RMU2026.05 | 56.8 | 1.204 | 33 | 64.3 | 31.4 | |
| NPOSAM=true2026.05 | 56 | 1.363 | 25.8 | 62.9 | 37.1 | |
| RMUSAM=true2026.05 | 54.9 | 1.201 | 34 | 65.6 | 31.6 |