Ring-a-bell
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Concept Removal | Ring-A-Bell Nudity | Attack Success Rate (ASR)95.6 | 52 | |
| Implicit Concept Erasure | Ring-A-Bell | ASR98 | 23 | |
| Concept Removal | Ring-A-Bell Violence | Attack Success Rate (ASR)88 | 22 | |
| Safety Unlearning Evaluation | Ring-A-Bell Violence (test) | ASR5.54 | 21 | |
| Safety Unlearning Evaluation | Ring-A-Bell Nudity (test) | ASR9.22 | 21 | |
| Concept Unlearning | Ring-A-Bell | Ring-A-Bell Score0.83 | 20 | |
| NSFW suppression | Ring-A-Bell | ASR1.3 | 18 | |
| Nudity Erasure | Ring-A-Bell | Generation Rate98.2 | 17 | |
| Defense against adversarial prompts | Ring-A-Bell (test) | Attack Success Rate0 | 15 | |
| Nudity Concept Removal | Ring-A-Bell 77 | Attack Success Rate (ASR)92.6 | 15 | |
| Adversarial Robustness in Concept Erasing | Ring-A-Bell K-16, K-38, K-77 | K-16 Score0.9579 | 14 | |
| Nudity erasure | Ring-A-Bell 1.5 (test) | Nudity Generation Rate17.9 | 13 | |
| Concept Erasure | Ring-A-Bell | ASR93 | 13 | |
| Safety Evaluation | Ring-A-Bell | Ring-16 Score4.93 | 13 | |
| Safe Text-to-Image Generation | Ring-A-Bell | ASR83.1 | 13 | |
| Nudity Concept Erasure | Ring-a-bell Adversarial Prompts | Erase Rate (%)100 | 13 | |
| Violence Concept Removal | Ring-A-Bell 5-77 | ASR99.2 | 12 | |
| Concept Unlearning Robustness | Ring-A-Bell adversarial prompts (K77) | ASR (Threshold 0.3)1.05 | 10 | |
| Concept Unlearning Robustness | Ring-A-Bell adversarial prompts K38 | ASR (T=0.3)1.05 | 10 | |
| Concept Unlearning Robustness | Ring-A-Bell adversarial prompts (K16) | ASR (T=0.3)71.58 | 10 | |
| Safe generation against nudity prompts | Ring-A-Bell | Attack Success Rate (ASR)5.1 | 9 | |
| Concept Erasure Robustness | Ring-A-Bell Union | Nudity Rate24.77 | 9 | |
| Concept Erasure Robustness | Ring-A-Bell | Nudity Rate11.01 | 9 | |
| Adversarial Attack | Ring-A-Bell (285 prompts) | ASR44.56 | 8 | |
| Text-to-Image Generation | Ring-a-bell (adversarial) | Q16 IP100 | 8 |
Showing 25 of 41 rows