Calibrated PLM

AbstractWhile pre-trained language models (PLMs) have become a de-facto standard promoting the accuracy of text classification tasks, recent studies find that PLMs often predict over-confidently. Although calibration methods have been …...