RobustBioNLP

Improving the robustness and performance of biomedical NLP models through adversarial training

In this repository you can find supplimentary materials and source codes for the paper titled "Improving the robustness and accuracy of biomedical language models through adversarial training". The paper can be found on arxive at (link to the arxive paper).

Biomedical/clinical NLP tasks

We used various biomedical/clinical text processing datasets covering five different NLP tasks.

Links to the datasets:

• BioText: https://biotext.berkeley.edu/
• MedNLI: https://physionet.org/content/mednli/1.0.0/
• MedSTS: https://n2c2.dbmi.hms.harvard.edu/track1
• PubMed-RCT: https://github.com/Franck-Dernoncourt/pubmed-rct
• PubMed-QA: https://pubmedqa.github.io/

Robust-BioMed-RoBERTa models

Here you can find links to the Robust-BioMede-RoBERTa models separately fine-tuned on adversarial samples for each of the biomedical/clinical NLP tasks:

• Robust-BioMed-RoBERTa-RelationClassification: https://huggingface.co/mmoradi/Robust-Biomed-RoBERTa-RelationClassification
• Robust-BioMed-RoBERTa-TextualInference: https://huggingface.co/mmoradi/Robust-Biomed-RxoBERTa-TextualInference
• Robust-BioMed-RoBERTa-SemanticSimilarity: https://huggingface.co/mmoradi/Robust-Biomed-RoBERTa-SemanticSimilarity
• Robust-BioMed-RoBERTa-TextClassification: https://huggingface.co/mmoradi/Robust-Biomed-RoBERTa-TextClassification
• Robust-BioMed-RoBERTa-QuestionAnswering: https://huggingface.co/mmoradi/Robust-Biomed-RoBERTa-QuestionAnswering