Natural Language Inference on SNLI (Adversarial - TextFooler-BERT)
62.3AccuracyInfoBERT
Evaluation results
| Method | Links | |
|---|---|---|
| InfoBERTTraining=Adversarial Training, Model=RoBERTa2020.10 | 62.3 | |
| FreeLBTraining=Adversarial Training, Model=RoBERTa2020.10 | 60.4 | |
| InfoBERTTraining=Standard Training, Model=RoBERTa2020.10 | 59.8 | |
| VanillaTraining=Standard Training, Model=RoBERTa2020.10 | 56.6 | |
| InfoBERTTraining=Adversarial Training, Model=BERT2020.10 | 50.8 | |
| FreeLBTraining=Adversarial Training, Model=BERT2020.10 | 46.6 | |
| InfoBERTTraining=Standard Training, Model=BERT2020.10 | 36.7 | |
| VanillaTraining=Standard Training, Model=BERT2020.10 | 0 |