Backdoor Detection on Perturbation-based Games Average
90AUROCSTRIP
Evaluation Results
| Method | Links | |
|---|---|---|
| STRIPStrategy=Original2026.06 | 90 | |
| PolicyGuardStrategy=Hard-coded2026.06 | 86.8 | |
| PolicyGuardStrategy=Original2026.06 | 85.6 | |
| SCALE-UPStrategy=Original2026.06 | 84.1 | |
| SHINEStrategy=Hard-coded2026.06 | 82.6 | |
| SHINEStrategy=Original2026.06 | 81.9 | |
| NCStrategy=Original2026.06 | 71.4 | |
| B3D (Black-box NC)Strategy=Hard-coded2026.06 | 50 | |
| STRIPStrategy=Hard-coded2026.06 | 48.2 | |
| SCALE-UPStrategy=Hard-coded2026.06 | 47.6 |