Text Classification on IMDB (test) (Adversarial Robustness Metrics)
3.6Attack Success Ratehoneypot defense
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| honeypot defenseVictim=RoBERTa_LARGE, Attack=AddWord2023.10 | 3.6 | — | — | 94.12 | |
| honeypot defenseVictim=RoBERTa_BASE, Attack=AddWord2023.10 | 5.6 | — | — | 93.72 | |
| honeypot defenseVictim=BERT_LARGE, Attack=AddWord2023.10 | 5.84 | — | — | 92.6 | |
| honeypot defenseVictim=RoBERTa_LARGE, Attack=AddSent2023.10 | 6.32 | — | — | 93.68 | |
| honeypot defenseVictim=RoBERTa_BASE, Attack=AddSent2023.10 | 6.56 | — | — | 92.72 | |
| honeypot defenseVictim=BERT_BASE, Attack=AddWord2023.10 | 7.2 | — | — | 91.32 | |
| honeypot defenseVictim=BERT_BASE, Attack=AddSent2023.10 | 8.16 | — | — | 91 | |
| honeypot defenseVictim=BERT_LARGE, Attack=AddSent2023.10 | 9.52 | — | — | 92.12 | |
| honeypot defenseVictim=RoBERTa_LARGE, Attack=Average2023.10 | 11.58 | — | — | 94.08 | |
| honeypot defenseVictim=RoBERTa_BASE, Attack=Average2023.10 | 13.56 | — | — | 93.19 | |
| honeypot defenseVictim=BERT_LARGE, Attack=Average2023.10 | 14.76 | — | — | 92.36 | |
| honeypot defenseVictim=BERT_BASE, Attack=Average2023.10 | 14.89 | — | — | 90.2 | |
| honeypot defenseVictim=RoBERTa_LARGE, Attack=StyleBKD2023.10 | 17.9 | — | — | 94.5 | |
| honeypot defenseVictim=RoBERTa_LARGE, Attack=SynBKD2023.10 | 18.4 | — | — | 94 | |
| honeypot defenseVictim=BERT_LARGE, Attack=StyleBKD2023.10 | 18.7 | — | — | 92.76 | |
| honeypot defenseVictim=RoBERTa_BASE, Attack=StyleBKD2023.10 | 19.36 | — | — | 93.12 | |
| honeypot defenseVictim=BERT_BASE, Attack=StyleBKD2023.10 | 20.6 | — | — | 89.44 | |
| honeypot defenseVictim=RoBERTa_BASE, Attack=SynBKD2023.10 | 22.7 | — | — | 93.2 | |
| honeypot defenseVictim=BERT_BASE, Attack=SynBKD2023.10 | 23.6 | — | — | 89.04 | |
| honeypot defenseVictim=BERT_LARGE, Attack=SynBKD2023.10 | 24.96 | — | — | 91.96 | |
| BERT-AttackVictim Model=BERT2023.05 | 88.2 | 5.3 | 0.78 | — | |
| SDM-ATTACKVictim Model=RoBERTa2023.05 | 88.3 | 8.3 | 0.7 | — | |
| TextFoolerVictim Model=BERT2023.05 | 88.7 | 7.6 | 0.76 | — | |
| SDM-ATTACKVictim Model=WordCNN2023.05 | 89.2 | 3.3 | 0.85 | — | |
| SDM-ATTACKVictim Model=WordLSTM2023.05 | 89.8 | 5.4 | 0.75 | — | |
| A2TVictim Model=BERT2023.05 | 89.9 | 4.4 | 0.79 | — | |
| SDM-ATTACKVictim Model=BERT2023.05 | 91.4 | 4.1 | 0.82 | — |