arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2203.01677cs.CLcs.CRcs.LG
收藏

文本分类中的词对抗性实例检测:基于鲁棒密度估计的Benchmark和Baseline

Detection of Word Adversarial Examples in Text Classification: Benchmark and Baseline via Robust Density Estimation

Seoul National University(首尔国立大学);NAVER WEBTOON AI(NAVER WEBTOON人工智能部门)

KiYoon Yoo, Jangho Kim, Jiho Jang, Nojun Kwak

摘要:

词级对抗攻击在NLP模型中取得了成功,近年来大大降低了基于transformer的模型的性能。作为一种对策,对抗性防御已被探索,但相对较少的努力已经检测到对抗性的例子。然而,检测对抗性示例对于自动化任务(例如,审查情感分析)可能是至关重要的,这些任务希望收集有关特定人群的信息,并且另外是迈向强大防御系统的一步。为此,我们在四个数据集和四个模型上发布了四种流行攻击方法的数据集,以鼓励在这一领域的进一步研究。与此同时,我们提出了一个基于密度估计的竞争性基线,该基线在30种攻击模型组合中的29种组合中具有最高的AUC。源代码可在https://github.com/anoymous92874838/text-adv-detection上获得。

英文摘要:

Word-level adversarial attacks have shown success in NLP models, drastically decreasing the performance of transformer-based models in recent years. As a countermeasure, adversarial defense has been explored, but relatively few efforts have been made to detect adversarial examples. However, detecting adversarial examples may be crucial for automated tasks (e.g. review sentiment analysis) that wish to amass information about a certain population and additionally be a step towards a robust defense system. To this end, we release a dataset for four popular attack methods on four datasets and four models to encourage further research in this field. Along with it, we propose a competitive baseline based on density estimation that has the highest AUC on 29 out of 30 dataset-attack-model combinations. Source code is available in https://github.com/anoymous92874838/text-adv-detection.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑