References & Citations
Computer Science > Computation and Language
Title: Adversarial Training with Fast Gradient Projection Method against Synonym Substitution based Text Attacks
(Submitted on 9 Aug 2020 (v1), last revised 16 Dec 2020 (this version, v4))
Abstract: Adversarial training is the most empirically successful approach in improving the robustness of deep neural networks for image classification.For text classification, however, existing synonym substitution based adversarial attacks are effective but not efficient to be incorporated into practical text adversarial training. Gradient-based attacks, which are very efficient for images, are hard to be implemented for synonym substitution based text attacks due to the lexical, grammatical and semantic constraints and the discrete text input space. Thereby, we propose a fast text adversarial attack method called Fast Gradient Projection Method (FGPM) based on synonym substitution, which is about 20 times faster than existing text attack methods and could achieve similar attack performance. We then incorporate FGPM with adversarial training and propose a text defense method called Adversarial Training with FGPM enhanced by Logit pairing (ATFL). Experiments show that ATFL could significantly improve the model robustness and block the transferability of adversarial examples.
Submission history
From: Xiaosen Wang [view email][v1] Sun, 9 Aug 2020 11:02:06 GMT (342kb,D)
[v2] Mon, 14 Sep 2020 14:59:24 GMT (144kb,D)
[v3] Wed, 9 Dec 2020 06:51:01 GMT (369kb,D)
[v4] Wed, 16 Dec 2020 03:01:25 GMT (365kb,D)
Link back to: arXiv, form interface, contact.