Natural Language Inference on adv-MNLI TextFooler-RoBERTa
65.6Accuracy (Matched)InfoBERT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| InfoBERTTraining=Adversarial Training, Model=BERT2020.10 | 65.6 | 67.3 | |
| FreeLBTraining=Adversarial Training, Model=BERT2020.10 | 64 | 62.9 | |
| InfoBERTTraining=Standard Training, Model=BERT2020.10 | 57.2 | 58.6 | |
| VanillaTraining=Standard Training, Model=BERT2020.10 | 57 | 57.5 | |
| InfoBERTTraining=Adversarial Training, Model=RoBERTa2020.10 | 56.9 | 55.5 | |
| InfoBERTTraining=Standard Training, Model=RoBERTa2020.10 | 50.3 | 52.1 | |
| FreeLBTraining=Adversarial Training, Model=RoBERTa2020.10 | 49.5 | 50.6 | |
| VanillaTraining=Standard Training, Model=RoBERTa2020.10 | 24.9 | 24.9 |