Text Classification on SST-2 (test) using ASR, CACC, and Delta Metrics
1.57Delta CACCAttDef
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AttDefAttacks=BadNLm2023.05 | 1.57 | — | — | 59.87 | |
| AttDefAttacks=Benign2023.05 | 1.68 | — | — | — | |
| AttDefAttacks=BadNLl2023.05 | 1.77 | — | — | 71.91 | |
| AttDefAttacks=Avg2023.05 | 1.77 | — | — | 50.72 | |
| AttDefAttacks=BadNLh2023.05 | 1.88 | — | — | 48.47 | |
| AttDefAttacks=InSent2023.05 | 1.97 | — | — | 22.63 | |
| ONIONAttacks=InSent2023.05 | 2.43 | — | — | 3.79 | |
| ONIONAttacks=Benign2023.05 | 2.6 | — | — | — | |
| ONIONAttacks=BadNLl2023.05 | 2.8 | — | — | 71.34 | |
| ONIONAttacks=Avg2023.05 | 2.85 | — | — | 44.86 | |
| ONIONAttacks=BadNLh2023.05 | 3.03 | — | — | 38.99 | |
| ONIONAttacks=BadNLm2023.05 | 3.14 | — | — | 65.33 | |
| AttDef w/o ELECTRAAttacks=BadNLm2023.05 | 5.64 | — | — | 67.7 | |
| AttDef w/o ELECTRAAttacks=Avg2023.05 | 7.39 | — | — | 56.73 | |
| AttDef w/o ELECTRAAttacks=InSent2023.05 | 7.58 | — | — | 28.4 | |
| AttDef w/o ELECTRAAttacks=Benign2023.05 | 7.73 | — | — | — | |
| AttDef w/o ELECTRAAttacks=BadNLl2023.05 | 7.9 | — | — | 82.68 | |
| AttDef w/o ELECTRAAttacks=BadNLh2023.05 | 8.12 | — | — | 48.13 | |
| Poisoned ModelAttacks=Benign2023.05 | — | — | 91.84 | — | |
| Poisoned ModelAttacks=BadNLl2023.05 | — | 99.93 | 91.31 | — | |
| Poisoned ModelAttacks=BadNLm2023.05 | — | 98.97 | 90.96 | — | |
| Poisoned ModelAttacks=BadNLh2023.05 | — | 89.78 | 90.87 | — | |
| Poisoned ModelAttacks=InSent2023.05 | — | 100 | 91.4 | — | |
| Poisoned ModelAttacks=Avg2023.05 | — | 97.13 | 91.17 | — |