This project explores using the effectiveness of self-ensembling for question answering systems. Obtaining a large training set can be necessary and challenging for building a question answering (QA) system. Self-ensembling has the potential to reduce the size of the training set and enable semi-supervised learning. As self-ensembling has shown promising results for visual domain adaptation, it is interesting to see if self-ensembling can improve the performance of a QA system.
The code is based on https://github.com/gregdurrett/nlp-qa-finalproj.git