Fair Generation on WinoBias Race (standard)
0.04Deviation RatioVector Interpret Diffusion
Evaluation results
| Method | Links | |
|---|---|---|
| Vector Interpret DiffusionAttribute=Teacher2025.03 | 0.04 | |
| Unified Concept EditingAttribute=Doctor2025.03 | 0.07 | |
| OursAttribute=Doctor2025.03 | 0.07 | |
| OursAttribute=Nurse2025.03 | 0.07 | |
| OursAttribute=Receptionist2025.03 | 0.07 | |
| OursAttribute=CEO2025.03 | 0.08 | |
| OursAttribute=Teacher2025.03 | 0.08 | |
| Unified Concept EditingAttribute=Receptionist2025.03 | 0.1 | |
| Unified Concept EditingAttribute=Teacher2025.03 | 0.1 | |
| Unified Concept EditingAttribute=CEO2025.03 | 0.13 | |
| Vector Interpret DiffusionAttribute=CEO2025.03 | 0.22 | |
| Unified Concept EditingAttribute=Nurse2025.03 | 0.25 | |
| Vector Interpret DiffusionAttribute=Doctor2025.03 | 0.26 | |
| Vector Interpret DiffusionAttribute=Nurse2025.03 | 0.3 | |
| Vector Interpret DiffusionAttribute=Receptionist2025.03 | 0.36 | |
| Standard Diffusion ModelAttribute=CEO2025.03 | 0.38 | |
| Standard Diffusion ModelAttribute=Teacher2025.03 | 0.51 | |
| Standard Diffusion ModelAttribute=Nurse2025.03 | 0.76 | |
| Standard Diffusion ModelAttribute=Receptionist2025.03 | 0.88 | |
| Standard Diffusion ModelAttribute=Doctor2025.03 | 0.92 |