Backdoor Defense on SST-2 (test)
-0.17ΔCACCBFClass
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BFClassAttack=BadNLm2023.05 | -0.17 | — | — | 51.51 | |
| BFClassAttack=Benign2023.05 | 0 | — | — | — | |
| BFClassAttack=InSent2023.05 | 0 | — | — | 0 | |
| BFClassAttack=Avg2023.05 | 0.02 | — | — | 38 | |
| BFClassAttack=BadNLl2023.05 | 0.13 | — | — | 87.08 | |
| BFClassAttack=BadNLh2023.05 | 0.14 | — | — | 13.44 | |
| AttDefAttack=Benign2023.05 | 1.84 | — | — | — | |
| AttDefAttack=BadNLl2023.05 | 1.92 | — | — | 80.24 | |
| AttDefAttack=BadNLm2023.05 | 1.92 | — | — | 67.24 | |
| AttDefAttack=Avg2023.05 | 2.08 | — | — | 64.64 | |
| AttDefAttack=InSent2023.05 | 2.22 | — | — | 57.08 | |
| AttDefAttack=BadNLh2023.05 | 2.49 | — | — | 53.99 | |
| Poisoned ModelAttack=Benign2023.05 | — | — | 91.84 | — | |
| Poisoned ModelAttack=BadNLl2023.05 | — | 99.93 | 91.31 | — | |
| Poisoned ModelAttack=BadNLm2023.05 | — | 98.97 | 90.96 | — | |
| Poisoned ModelAttack=BadNLh2023.05 | — | 89.78 | 90.87 | — | |
| Poisoned ModelAttack=InSent2023.05 | — | 100 | 91.4 | — | |
| Poisoned ModelAttack=Avg2023.05 | — | 97.17 | 91.13 | — |