Natural Language Inference on adv-MNLI TextFooler-BERT
73.2Accuracy (Matched)InfoBERT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| InfoBERTTraining=Adversarial Training, Model=RoBERTa2020.10 | 73.2 | 73.1 | |
| FreeLBTraining=Adversarial Training, Model=RoBERTa2020.10 | 70.3 | 72.1 | |
| InfoBERTTraining=Standard Training, Model=RoBERTa2020.10 | 69.8 | 70.6 | |
| VanillaTraining=Standard Training, Model=RoBERTa2020.10 | 68.1 | 68.6 | |
| InfoBERTTraining=Adversarial Training, Model=BERT2020.10 | 61.3 | 62.7 | |
| FreeLBTraining=Adversarial Training, Model=BERT2020.10 | 60 | 60.7 | |
| InfoBERTTraining=Standard Training, Model=BERT2020.10 | 43.5 | 46.6 | |
| VanillaTraining=Standard Training, Model=BERT2020.10 | 0 | 0 |