Defending Pre-trained Language Models from Adversarial Word Substitution Without Performance Sacrifice.

Rongzhou Bao, Jiayi Wang, Hai Zhao

Browse the full ACL paper archive.