Textual Adversarial Attacking with Limited Queries

Recent studies have shown that natural language processing (NLP) models are vulnerable to adversarial examples, which are maliciously designed by adding small perturbations to benign inputs that are imperceptible to the human eye, leading to false predictions by the target model. Compared to charact...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Yu Zhang, Junan Yang, Xiaoshuai Li, Hui Liu, Kun Shao
Formato:	article
Lenguaje:	EN
Publicado:	MDPI AG 2021
Materias:	machine learning adversarial attack natural language processing (NLP) black box Electronics TK7800-8360
Acceso en línea:	https://doaj.org/article/89ac0f34923e4dbbb9b19901a365a476
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:89ac0f34923e4dbbb9b19901a365a476
record_format	dspace
spelling	oai:doaj.org-article:89ac0f34923e4dbbb9b19901a365a4762021-11-11T15:39:50ZTextual Adversarial Attacking with Limited Queries10.3390/electronics102126712079-9292https://doaj.org/article/89ac0f34923e4dbbb9b19901a365a4762021-10-01T00:00:00Zhttps://www.mdpi.com/2079-9292/10/21/2671https://doaj.org/toc/2079-9292Recent studies have shown that natural language processing (NLP) models are vulnerable to adversarial examples, which are maliciously designed by adding small perturbations to benign inputs that are imperceptible to the human eye, leading to false predictions by the target model. Compared to character- and sentence-level textual adversarial attacks, word-level attack can generate higher-quality adversarial examples, especially in a black-box setting. However, existing attack methods usually require a huge number of queries to successfully deceive the target model, which is costly in a real adversarial scenario. Hence, finding appropriate models is difficult. Therefore, we propose a novel attack method, the main idea of which is to fully utilize the adversarial examples generated by the local model and transfer part of the attack to the local model to complete ahead of time, thereby reducing costs related to attacking the target model. Extensive experiments conducted on three public benchmarks show that our attack method can not only improve the success rate but also reduce the cost, while outperforming the baselines by a significant margin.Yu ZhangJunan YangXiaoshuai LiHui LiuKun ShaoMDPI AGarticlemachine learningadversarial attacknatural language processing (NLP)black boxElectronicsTK7800-8360ENElectronics, Vol 10, Iss 2671, p 2671 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	machine learning adversarial attack natural language processing (NLP) black box Electronics TK7800-8360
spellingShingle	machine learning adversarial attack natural language processing (NLP) black box Electronics TK7800-8360 Yu Zhang Junan Yang Xiaoshuai Li Hui Liu Kun Shao Textual Adversarial Attacking with Limited Queries
description	Recent studies have shown that natural language processing (NLP) models are vulnerable to adversarial examples, which are maliciously designed by adding small perturbations to benign inputs that are imperceptible to the human eye, leading to false predictions by the target model. Compared to character- and sentence-level textual adversarial attacks, word-level attack can generate higher-quality adversarial examples, especially in a black-box setting. However, existing attack methods usually require a huge number of queries to successfully deceive the target model, which is costly in a real adversarial scenario. Hence, finding appropriate models is difficult. Therefore, we propose a novel attack method, the main idea of which is to fully utilize the adversarial examples generated by the local model and transfer part of the attack to the local model to complete ahead of time, thereby reducing costs related to attacking the target model. Extensive experiments conducted on three public benchmarks show that our attack method can not only improve the success rate but also reduce the cost, while outperforming the baselines by a significant margin.
format	article
author	Yu Zhang Junan Yang Xiaoshuai Li Hui Liu Kun Shao
author_facet	Yu Zhang Junan Yang Xiaoshuai Li Hui Liu Kun Shao
author_sort	Yu Zhang
title	Textual Adversarial Attacking with Limited Queries
title_short	Textual Adversarial Attacking with Limited Queries
title_full	Textual Adversarial Attacking with Limited Queries
title_fullStr	Textual Adversarial Attacking with Limited Queries
title_full_unstemmed	Textual Adversarial Attacking with Limited Queries
title_sort	textual adversarial attacking with limited queries
publisher	MDPI AG
publishDate	2021
url	https://doaj.org/article/89ac0f34923e4dbbb9b19901a365a476
work_keys_str_mv	AT yuzhang textualadversarialattackingwithlimitedqueries AT junanyang textualadversarialattackingwithlimitedqueries AT xiaoshuaili textualadversarialattackingwithlimitedqueries AT huiliu textualadversarialattackingwithlimitedqueries AT kunshao textualadversarialattackingwithlimitedqueries
_version_	1718434544940482560

Textual Adversarial Attacking with Limited Queries

Ejemplares similares