000264189 001__ 264189
000264189 005__ 20190228110613.0
000264189 037__ $$aCONF
000264189 245__ $$aUniversal Adversarial Attacks on Text Classifiers
000264189 260__ $$c2019
000264189 269__ $$a2019
000264189 336__ $$aConference Papers
000264189 520__ $$aDespite the vast success neural networks have achieved in different application domains, they have been proven to be vulnerable to adversarial perturbations (small changes in the input), which lead them to produce the wrong output. In this paper, we propose a novel method, based on gradient projection, for generating universal adversarial perturbations for text; namely sequence of words that can be added to any input in order to fool the classifier with high probability. We observed that text classifiers are quite vulnerable to such perturbations: inserting even a single adversarial word to the beginning of every input sequence can drop the accuracy from 93% to 50%.
000264189 700__ $$aBehjati, Melika
000264189 700__ $$0249003$$aMoosavi Dezfooli, Seyed Mohsen
000264189 700__ $$aSoleymani Baghshah, Mahdieh
000264189 700__ $$0241061$$aFrossard, Pascal
000264189 7112_ $$aInternational Conference on Acoustics, Speech, and Signal Processing (ICASSP)$$cBrighton, UK$$d2019
000264189 8560_ $$fseyed.moosavi@epfl.ch
000264189 909C0 $$zMarselli, Béatrice$$xU10851$$pLTS4$$mpascal.frossard@epfl.ch$$0252393
000264189 909CO $$ooai:infoscience.epfl.ch:264189$$pSTI$$pconf
000264189 960__ $$aseyed.moosavi@epfl.ch
000264189 961__ $$afantin.reichler@epfl.ch
000264189 973__ $$aOTHER$$rREVIEWED
000264189 980__ $$aCONF
000264189 981__ $$aoverwrite