Sentence-pair classification on QNLI
92.45AccuracyLeRaC
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LeRaCTraining Regime=LeRaC, Model=BERT_large2022.05 | 92.45 | — | |
| conventionalTraining Regime=conventional, Model=BERT_large2022.05 | 92.13 | — | |
| BERT (Full Clean Training)Training Data=Full clean dataset2022.10 | 91.56 | — | |
| Backdoored Model (No Defense)Backdoor Attack=Word+EP, Clean training instances=642022.10 | 91.56 | 96.23 | |
| CBSTraining Regime=CBS, Model=BERT_large2022.05 | 91.47 | — | |
| Backdoored Model (No Defense)Backdoor Attack=Trigger Word, Clean training instances=642022.10 | 90.79 | 99.98 | |
| Fine-mixingBackdoor Attack=Sentence (Scratch), Clean training instances=642022.10 | 84.29 | 86.02 | |
| Fine-mixingBackdoor Attack=Word (Scratch), Clean training instances=642022.10 | 84.07 | 30.68 | |
| Fine-mixingBackdoor Attack=Word+EP, Clean training instances=642022.10 | 82.83 | 29.52 | |
| Fine-mixingBackdoor Attack=Trigger Sentence, Clean training instances=642022.10 | 82.83 | 31.4 | |
| Fine-mixingBackdoor Attack=Sentence (Scratch), Clean training instances=1282022.10 | 82.47 | 77.23 | |
| Fine-mixingBackdoor Attack=Sentence (Scratch), Clean training instances=2562022.10 | 81.9 | 60.74 | |
| Fine-mixingBackdoor Attack=Trigger Word, Clean training instances=642022.10 | 81.68 | 21.77 | |
| Fine-mixingBackdoor Attack=Word (Scratch), Clean training instances=1282022.10 | 81.37 | 22.73 | |
| Fine-mixingBackdoor Attack=Sentence (Scratch), Clean training instances=5122022.10 | 80.3 | 21.85 | |
| Initial BERT (Clean fine-tuned)Clean training instances=1282022.10 | 67.27 | — | |
| LeRaCTraining Regime=LeRaC, Model=LSTM2022.05 | 59.98 | — | |
| CBSTraining Regime=CBS, Model=LSTM2022.05 | 59.89 | — | |
| conventionalTraining Regime=conventional, Model=LSTM2022.05 | 59.42 | — | |
| Initial BERT (Clean fine-tuned)Clean training instances=642022.10 | 49.95 | — |