Natural Language Inference on adv-SNLI TextFooler-RoBERTa
52.6AccuracyInfoBERT
Evaluation results
| Method | Links | |
|---|---|---|
| InfoBERTTraining=Adversarial Training, Model=BERT2020.10 | 52.6 | |
| FreeLBTraining=Adversarial Training, Model=BERT2020.10 | 50.5 | |
| InfoBERTTraining=Standard Training, Model=BERT2020.10 | 45.4 | |
| VanillaTraining=Standard Training, Model=BERT2020.10 | 44.9 | |
| InfoBERTTraining=Adversarial Training, Model=RoBERTa2020.10 | 43.4 | |
| InfoBERTTraining=Standard Training, Model=RoBERTa2020.10 | 42.5 | |
| FreeLBTraining=Adversarial Training, Model=RoBERTa2020.10 | 41.2 | |
| VanillaTraining=Standard Training, Model=RoBERTa2020.10 | 19.4 |