Concept Unlearning Robustness on Ring-A-Bell adversarial prompts (K16)
71.58ASR (T=0.3)ESD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ESD2026.03 | 71.58 | 61.05 | 46.32 | 18.95 | |
| AP2026.03 | 56.84 | 46.32 | 32.63 | 14.74 | |
| Diversified-APstrategy=Diversified2026.03 | 51.58 | 38.95 | 25.26 | 10.53 | |
| Diversified-AGEstrategy=Diversified2026.03 | 51.58 | 40 | 16.84 | 4.21 | |
| UCE2026.03 | 49.47 | 41.05 | 15.79 | 4.21 | |
| Diversified-ESDstrategy=Diversified2026.03 | 40 | 29.47 | 20 | 5.26 | |
| AGE2026.03 | 35.79 | 32.63 | 17.89 | 5.26 | |
| Diversified-UCEstrategy=Diversified2026.03 | 27.37 | 22.11 | 14.74 | 3.16 | |
| ACE2026.03 | 3.16 | 0 | 0 | 0 | |
| Diversified-ACEstrategy=Diversified2026.03 | 0 | 0 | 0 | 0 |