Bad Characters: Imperceptible NLP Attacks

Author:

Boucher Nicholas1,Shumailov Ilia2,Anderson Ross3,Papernot Nicolas2

Affiliation:

1. University of Cambridgem, Computer Science & Technology

2. University of Cambridge and Vector Institute

3. University of Cambridge and University of Edinburgh

Publisher

IEEE

Reference84 articles.

1. Roberta: A robustly optimized BERT pretraining approach;liu;CoRR,2019

2. Perspective API;jigsaw,2021

3. Wikipedia talk labels: Toxicity;thain,2017

4. Toxic comment classifier,2020

5. Münchener Digitalisierungszentrum (MDZ) - Bayerische Staatsbibliothek;BERT Large Cased Finetuned CoNLL03 English,2020

Cited by 25 articles. 订阅此论文施引文献 订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献

1. A Survey of Text Watermarking in the Era of Large Language Models;ACM Computing Surveys;2024-09-03

2. The Impact of Uniform Inputs on Activation Sparsity and Energy-Latency Attacks in Computer Vision;2024 IEEE Security and Privacy Workshops (SPW);2024-05-23

3. Employing Invisible Unicode Characters for Preserving Privacy and Enhancing Communication Security;2024 21st International Multi-Conference on Systems, Signals & Devices (SSD);2024-04-22

4. ChatGPT-Resistant Screening Instrument for Identifying Non-Programmers;Proceedings of the IEEE/ACM 46th International Conference on Software Engineering;2024-04-12

5. Understanding, Uncovering, and Mitigating the Causes of Inference Slowdown for Language Models;2024 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML);2024-04-09

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3