Impact of Adversarial Training on Robustness and Generalizability of Language Models

Enes Altinisik, Hassan Sajjad, Husrev Sencar, Safa Messaoud, Sanjay Chawla


Abstract
Adversarial training is widely acknowledged as the most effective defense against adversarial attacks. However, it is also well established that achieving both robustness and generalization in adversarially trained models involves a trade-off. The goal of this work is to provide an in depth comparison of different approaches for adversarial training in language models. Specifically, we study the effect of pre-training data augmentation as well as training time input perturbations vs. embedding space perturbations on the robustness and generalization of transformer-based language models. Our findings suggest that better robustness can be achieved by pre-training data augmentation or by training with input space perturbation. However, training with embedding space perturbation significantly improves generalization. A linguistic correlation analysis of neurons of the learned models reveal that the improved generalization is due to ‘more specialized’ neurons. To the best of our knowledge, this is the first work to carry out a deep qualitative analysis of different methods of generating adversarial examples in adversarial training of language models.
Anthology ID:
2023.findings-acl.496
Volume:
Findings of the Association for Computational Linguistics: ACL 2023
Month:
July
Year:
2023
Address:
Toronto, Canada
Editors:
Anna Rogers, Jordan Boyd-Graber, Naoaki Okazaki
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
7828–7840
Language:
URL:
https://aclanthology.org/2023.findings-acl.496
DOI:
10.18653/v1/2023.findings-acl.496
Bibkey:
Cite (ACL):
Enes Altinisik, Hassan Sajjad, Husrev Sencar, Safa Messaoud, and Sanjay Chawla. 2023. Impact of Adversarial Training on Robustness and Generalizability of Language Models. In Findings of the Association for Computational Linguistics: ACL 2023, pages 7828–7840, Toronto, Canada. Association for Computational Linguistics.
Cite (Informal):
Impact of Adversarial Training on Robustness and Generalizability of Language Models (Altinisik et al., Findings 2023)
Copy Citation:
PDF:
https://aclanthology.org/2023.findings-acl.496.pdf
Video:
 https://aclanthology.org/2023.findings-acl.496.mp4