Dual-Targeted Textfooler Attack on Text Classification Systems
نویسندگان
چکیده
Deep neural networks provide good performance on classification tasks such as those for image, audio, and text classification. However, are vulnerable to adversarial examples. An example is a sample created by adding small noise an original data in way that it will be correctly classified human but misclassified deep network. Studies examples have focused mainly the image field, research expanding into field well. Adversarial designed with two targets mind can useful certain situations. In military scenario, example, if enemy models A B use recognition model, may desirable cause model tanks go right self-propelled guns left using strategically messages. Such dual-targeted could accomplish this causing different misclassifications models, contrast single-target produced existing methods. paper, I propose method creating textual attacking system. Unlike methods, which images, proposed creates class each of while maintaining meaning grammar sentence, substituting words importance. Experiments were conducted SNLI dataset TensorFlow library. The results demonstrate generate average attack success rate 82.2% models.
منابع مشابه
Cipher text only attack on speech time scrambling systems using correction of audio spectrogram
Recently permutation multimedia ciphers were broken in a chosen-plaintext scenario. That attack models a very resourceful adversary which may not always be the case. To show insecurity of these ciphers, we present a cipher-text only attack on speech permutation ciphers. We show inherent redundancies of speech can pave the path for a successful cipher-text only attack. To that end, regularities ...
متن کاملBirthday Attack on Dual EWCDM
In CRYPTO 2017, Mennink and Neves showed almost nbit security for a dual version of EWCDM. In this paper we describe a birthday attack on this construction which violates their claim.
متن کاملA Simple Generic Attack on Text Captchas
Text-based Captchas have been widely deployed across the Internet to defend against undesirable or malicious bot programs. Many attacks have been proposed; these fine prior art advanced the scientific understanding of Captcha robustness, but most of them have a limited applicability. In this paper, we report a simple, low-cost but powerful attack that effectively breaks a wide range of text Cap...
متن کاملA Novel Field Learning Algorithm for Dual Imbalance Text Classification
Fish-net algorithm is a novel field learning algorithm which derives classification rules by looking at the range of values of each attribute instead of the individual point values. In this paper, we present a Feature Selection Fish-net learning algorithm to solve the Dual Imbalance problem on text classification. Dual imbalance includes the instance imbalance and feature imbalance. The instanc...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: IEEE Access
سال: 2023
ISSN: ['2169-3536']
DOI: https://doi.org/10.1109/access.2021.3121366