Trigger Search on Backdoored Models Word-Level BERT-based (test)
86.32Recall (NeuBA)Patronus
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PatronusFuzzy Search Rounds (FS)=5, Optimization Strategy=beam search2025.12 | 86.32 | 14.92 | 77.01 | 31.25 | 83.63 | 31.56 | |
| PatronusFuzzy Search Rounds (FS)=3, Optimization Strategy=beam search2025.12 | 78.95 | 8.89 | 71.26 | 18.6 | 74.85 | 18.49 | |
| Patronus-greedyFuzzy Search Rounds (FS)=5, Optimization Strategy=greedy optimization2025.12 | 77.89 | 3.23 | 58.62 | 4.74 | 64.91 | 4.72 | |
| PatronusFuzzy Search Rounds (FS)=2, Optimization Strategy=beam search2025.12 | 76.84 | 5.86 | 58.05 | 14.29 | 69.01 | 11.44 | |
| LMSanitatorFuzzy Search Rounds (FS)=1002025.12 | 75.79 | 23.69 | 60.92 | 60.21 | 70.18 | 48.55 | |
| LMSanitatorFuzzy Search Rounds (FS)=502025.12 | 71.58 | 10.85 | 50.57 | 31.94 | 61.99 | 36.52 | |
| Patronus-greedyFuzzy Search Rounds (FS)=3, Optimization Strategy=greedy optimization2025.12 | 69.47 | 1.4 | 45.4 | 2.46 | 56.14 | 3.62 | |
| LMSanitatorFuzzy Search Rounds (FS)=252025.12 | 66.32 | 4.81 | 41.95 | 9.69 | 53.22 | 9.35 | |
| Patronus-greedyFuzzy Search Rounds (FS)=2, Optimization Strategy=greedy optimization2025.12 | 64.21 | 0.95 | 38.51 | 1.99 | 48.54 | 1.73 |