Sequential Model Defense on Gemma-S Case II: Base -> Unbias -> CAT
2.98Fairness ScoreUnbias
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| UnbiasDefense=+D12026.05 | 2.98 | 66.1 | |
| Conflict-Guided Layer FreezingDefense=+D1+D2 (frozen L9)2026.05 | 6.56 | 29.8 | |
| Unbias -> CATDefense=+D1+D22026.05 | 8.08 | 31.9 | |
| BaseDefense=Base2026.05 | 14.88 | 55.9 |