We propose the first general-purpose gradient-based attack against transformer models. Instead of searching for a single adversarial example, we search for a distribution of adversarial examples parameterized by a continuous-valued matrix, hence enabling gradient-based optimization. We empirically demonstrate that our white-box attack attains state-of-the-art attack performance on a variety of natural language tasks. Furthermore, we show that a powerful black-box transfer attack, enabled by sampling from the adversarial distribution, matches or exceeds existing methods, while only requiring hard-label outputs.
@article{arxiv.2104.13733,
title = {Gradient-based Adversarial Attacks against Text Transformers},
author = {Chuan Guo and Alexandre Sablayrolles and Hervé Jégou and Douwe Kiela},
journal= {arXiv preprint arXiv:2104.13733},
year = {2021}
}