Backdoor Defense on OLID (test)
2.21ΔCACCAttDef
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AttDefAttack=BadNLm2023.05 | 2.21 | — | — | 61.87 | |
| AttDefAttack=Benign2023.05 | 1.65 | — | — | — | |
| AttDefAttack=BadNLh2023.05 | 1.33 | — | — | 77.9 | |
| AttDefAttack=Avg2023.05 | 1.24 | — | — | 71.08 | |
| AttDefAttack=BadNLl2023.05 | 0.77 | — | — | 79.61 | |
| AttDefAttack=InSent2023.05 | 0.26 | — | — | 64.94 | |
| BFClassAttack=BadNLh2023.05 | 0.16 | — | — | 82.5 | |
| BFClassAttack=BadNLl2023.05 | -0.23 | — | — | 46.58 | |
| BFClassAttack=BadNLm2023.05 | -0.39 | — | — | 45.45 | |
| BFClassAttack=Avg2023.05 | -0.56 | — | — | 43.63 | |
| BFClassAttack=Benign2023.05 | -0.82 | — | — | — | |
| BFClassAttack=InSent2023.05 | -1.51 | — | — | 0 | |
| Poisoned ModelAttack=Benign2023.05 | — | — | 81.82 | — | |
| Poisoned ModelAttack=BadNLl2023.05 | — | 100 | 81.23 | — | |
| Poisoned ModelAttack=BadNLm2023.05 | — | 100 | 81.3 | — | |
| Poisoned ModelAttack=BadNLh2023.05 | — | 97.19 | 81.42 | — | |
| Poisoned ModelAttack=InSent2023.05 | — | 100 | 80.91 | — | |
| Poisoned ModelAttack=Avg2023.05 | — | 99.3 | 81.22 | — |