Improving Robustness of Language Models from a Geometry-aware Perspective

Bin Zhu, Zhaoquan Gu, Le Wang, Jinyin Chen, Qi Xuan
2022 Findings of the Association for Computational Linguistics: ACL 2022   unpublished
Recent studies have found that removing the norm-bounded projection and increasing search steps in adversarial training can significantly improve robustness. However, we observe that a too large number of search steps can hurt accuracy. We aim to obtain strong robustness efficiently using fewer steps. Through a toy experiment, we find that perturbing the clean data to the decision boundary but not crossing it does not degrade the test accuracy. Inspired by this, we propose friendly adversarial
more » ... ata augmentation (FADA) to generate friendly adversarial data. On top of FADA, we propose geometry-aware adversarial training (GAT) to perform adversarial training on friendly adversarial data so that we can save a large number of search steps. Comprehensive experiments across two widely used datasets and three pretrained language models demonstrate that GAT can obtain stronger robustness via fewer steps. In addition, we provide extensive empirical results and in-depth analyses on robustness to facilitate future studies.
doi:10.18653/v1/2022.findings-acl.246 fatcat:ktage4iryfg5thkzhkucstzbq4