Concept Unlearning Robustness on Ring-A-Bell adversarial prompts (K77)
1.05ASR (Threshold 0.3)ACE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ACE2026.03 | 1.05 | 0 | 0 | 0 | |
| Diversified-ACEstrategy=Diversified2026.03 | 1.05 | 1.05 | 0 | 0 | |
| Diversified-UCEstrategy=Diversified2026.03 | 22.11 | 15.79 | 7.37 | 1.05 | |
| Diversified-ESDstrategy=Diversified2026.03 | 36.84 | 22.21 | 15.79 | 6.32 | |
| AGE2026.03 | 41.05 | 31.58 | 17.89 | 8.42 | |
| UCE2026.03 | 49.47 | 41.05 | 18.95 | 2.11 | |
| Diversified-APstrategy=Diversified2026.03 | 49.47 | 38.95 | 22.11 | 9.47 | |
| Diversified-AGEstrategy=Diversified2026.03 | 53.68 | 41.05 | 17.89 | 4.21 | |
| AP2026.03 | 65.26 | 54.47 | 33.68 | 14.74 | |
| ESD2026.03 | 74.74 | 66.32 | 45.26 | 25.26 |