Adversarial Attack on MGT Detector Evaluation Set XLM-RoBERTa-Base (test)
96.58ASR (%)GREATER-A
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GREATER-AAttack Type=Query-based2025.02 | 96.58 | 62.63 | 7.26 | 35.22 | 0.9506 | 9.21 | |
| HOAAttack Type=Query-based2025.02 | 88.13 | 283.89 | 23.57 | 102.87 | 0.8854 | 72.16 | |
| ABPAttack Type=Query-based2025.02 | 75.65 | 785.18 | 14.63 | 39.61 | 0.8709 | 26.4 | |
| TextFoolerAttack Type=Query-based2025.02 | 72.29 | 690.92 | 6.26 | 46.89 | 0.9302 | 21.07 | |
| BERTAttackAttack Type=Query-based2025.02 | 71.49 | 411.54 | 5.79 | 36.15 | 0.9402 | 15.78 | |
| GREATER-AAttack Type=Zero-query2025.02 | 69.11 | — | — | 43.11 | 0.92 | 4.55 | |
| FastTextDodgerAttack Type=Query-based2025.02 | 63.78 | 745.75 | 13.29 | 76.15 | 0.9188 | 55.14 | |
| T-PGDAttack Type=Zero-query2025.02 | 62.4 | — | — | 140.13 | 0.82 | 44.99 | |
| PWWSAttack Type=Query-based2025.02 | 61.45 | 1,197.95 | 4.71 | 37.85 | 0.9488 | 12.76 | |
| T-PGDAttack Type=Query-based2025.02 | 49.9 | 354.75 | 38.01 | 181.31 | 0.8197 | 35.09 | |
| A2T (White Box)Attack Type=Query-based2025.02 | 45.47 | 293.26 | 7.01 | 62.84 | 0.9215 | 32.07 | |
| WordNetAttack Type=Zero-query2025.02 | 42.6 | — | — | 26.27 | 0.9 | 4.26 | |
| RewriteAttack Type=Zero-query2025.02 | 36.47 | — | — | 92.08 | 0.79 | 15.62 | |
| HMGCAttack Type=Zero-query2025.02 | 30 | — | — | 12.94 | 0.84 | 36.9 | |
| Back TranslationAttack Type=Zero-query2025.02 | 2.4 | — | — | 6.4 | 0.91 | 3.06 |