Step by Step Loss Goes Very Far: Multi-Step Quantization for Adversarial Text Attacks

Piotr Gaiński, Klaudia Bałazy


Abstract
We propose a novel gradient-based attack against transformer-based language models that searches for an adversarial example in a continuous space of tokens probabilities. Our algorithm mitigates the gap between adversarial loss for continuous and discrete text representations by performing multi-step quantization in a quantization-compensation loop. Experiments show that our method significantly outperforms other approaches on various natural language processing (NLP) tasks.
Anthology ID:
2023.eacl-main.149
Volume:
Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics
Month:
May
Year:
2023
Address:
Dubrovnik, Croatia
Editors:
Andreas Vlachos, Isabelle Augenstein
Venue:
EACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
2038–2048
Language:
URL:
https://aclanthology.org/2023.eacl-main.149
DOI:
10.18653/v1/2023.eacl-main.149
Bibkey:
Cite (ACL):
Piotr Gaiński and Klaudia Bałazy. 2023. Step by Step Loss Goes Very Far: Multi-Step Quantization for Adversarial Text Attacks. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 2038–2048, Dubrovnik, Croatia. Association for Computational Linguistics.
Cite (Informal):
Step by Step Loss Goes Very Far: Multi-Step Quantization for Adversarial Text Attacks (Gaiński & Bałazy, EACL 2023)
Copy Citation:
PDF:
https://aclanthology.org/2023.eacl-main.149.pdf
Video:
 https://aclanthology.org/2023.eacl-main.149.mp4