Question Answering on SQuAD AddSent 1.0 (adversarial)
78.5F1 ScoreInfoBERT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| InfoBERTTraining protocol=Standard Training, Model backbone=RoBERTa_Large2020.10 | 78.5 | 72.9 | |
| InfoBERTTraining protocol=Adversarial Training, Model backbone=RoBERTa_Large2020.10 | 78 | 71.8 | |
| FreeLBTraining protocol=Adversarial Training, Model backbone=RoBERTa_Large2020.10 | 76.3 | 70.3 | |
| ALUMTraining protocol=Adversarial Training, Model backbone=RoBERTa_Large2020.10 | 75.5 | 69.4 | |
| VanillaTraining protocol=Standard Training, Model backbone=RoBERTa_Large2020.10 | 72.9 | 66.6 |