Text Classification on IMDB (test) (ASR and Delta Metrics)
2.07Delta CACCAttDef w/o ELECTRA
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AttDef w/o ELECTRAAttacks=Benign2023.05 | 2.07 | — | — | — | |
| AttDef w/o ELECTRAAttacks=BadNLm2023.05 | 2.02 | — | — | 50.32 | |
| AttDefAttacks=Benign2023.05 | 2.02 | — | — | — | |
| AttDefAttacks=BadNLm2023.05 | 1.96 | — | — | 51.51 | |
| AttDef w/o ELECTRAAttacks=InSent2023.05 | 1.93 | — | — | 88.45 | |
| AttDef w/o ELECTRAAttacks=Avg2023.05 | 1.87 | — | — | 50.45 | |
| AttDefAttacks=InSent2023.05 | 1.86 | — | — | 87.44 | |
| AttDefAttacks=Avg2023.05 | 1.86 | — | — | 51.33 | |
| AttDef w/o ELECTRAAttacks=BadNLh2023.05 | 1.78 | — | — | 43.66 | |
| AttDefAttacks=BadNLh2023.05 | 1.76 | — | — | 45.54 | |
| AttDef w/o ELECTRAAttacks=BadNLl2023.05 | 1.71 | — | — | 19.39 | |
| AttDefAttacks=BadNLl2023.05 | 1.7 | — | — | 20.84 | |
| ONIONAttacks=InSent2023.05 | 0.39 | — | — | 0.19 | |
| ONIONAttacks=BadNLh2023.05 | 0.35 | — | — | 0.08 | |
| ONIONAttacks=Avg2023.05 | 0.33 | — | — | 0.14 | |
| ONIONAttacks=BadNLm2023.05 | 0.31 | — | — | 0.1 | |
| ONIONAttacks=Benign2023.05 | 0.3 | — | — | — | |
| ONIONAttacks=BadNLl2023.05 | 0.27 | — | — | 0.18 | |
| Poisoned ModelAttacks=Benign2023.05 | — | — | 93.84 | — | |
| Poisoned ModelAttacks=BadNLl2023.05 | — | 98.99 | 93.86 | — | |
| Poisoned ModelAttacks=BadNLm2023.05 | — | 99.96 | 93.82 | — | |
| Poisoned ModelAttacks=BadNLh2023.05 | — | 98.74 | 93.76 | — | |
| Poisoned ModelAttacks=InSent2023.05 | — | 97.73 | 92.7 | — | |
| Poisoned ModelAttacks=Avg2023.05 | — | 99.36 | 93.78 | — |