Adversarial Example Defenses: Ensembles of Weak Defenses are not Strong. He, W., Wei, J., Chen, X., Carlini, N., & Song, D. arXiv:1706.04701 [cs], June, 2017. arXiv: 1706.04701![Adversarial Example Defenses: Ensembles of Weak Defenses are not Strong [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper abstract bibtex Ongoing research has proposed several methods to defend neural networks against adversarial examples, many of which researchers have shown to be ineffective. We ask whether a strong defense can be created by combining multiple (possibly weak) defenses. To answer this question, we study three defenses that follow this approach. Two of these are recently proposed defenses that intentionally combine components designed to work well together. A third defense combines three independent defenses. For all the components of these defenses and the combined defenses themselves, we show that an adaptive adversary can create adversarial examples successfully with low distortion. Thus, our work implies that ensemble of weak defenses is not sufficient to provide strong defense against adversarial examples.
@article{he_adversarial_2017,
title = {Adversarial {Example} {Defenses}: {Ensembles} of {Weak} {Defenses} are not {Strong}},
shorttitle = {Adversarial {Example} {Defenses}},
url = {http://arxiv.org/abs/1706.04701},
abstract = {Ongoing research has proposed several methods to defend neural networks against adversarial examples, many of which researchers have shown to be ineffective. We ask whether a strong defense can be created by combining multiple (possibly weak) defenses. To answer this question, we study three defenses that follow this approach. Two of these are recently proposed defenses that intentionally combine components designed to work well together. A third defense combines three independent defenses. For all the components of these defenses and the combined defenses themselves, we show that an adaptive adversary can create adversarial examples successfully with low distortion. Thus, our work implies that ensemble of weak defenses is not sufficient to provide strong defense against adversarial examples.},
journal = {arXiv:1706.04701 [cs]},
author = {He, Warren and Wei, James and Chen, Xinyun and Carlini, Nicholas and Song, Dawn},
month = jun,
year = {2017},
note = {arXiv: 1706.04701},
keywords = {⛔ No DOI found}
}
Downloads: 0
{"_id":"KwpSuRvSktGxWSNwr","bibbaseid":"he-wei-chen-carlini-song-adversarialexampledefensesensemblesofweakdefensesarenotstrong-2017","authorIDs":[],"author_short":["He, W.","Wei, J.","Chen, X.","Carlini, N.","Song, D."],"bibdata":{"bibtype":"article","type":"article","title":"Adversarial Example Defenses: Ensembles of Weak Defenses are not Strong","shorttitle":"Adversarial Example Defenses","url":"http://arxiv.org/abs/1706.04701","abstract":"Ongoing research has proposed several methods to defend neural networks against adversarial examples, many of which researchers have shown to be ineffective. We ask whether a strong defense can be created by combining multiple (possibly weak) defenses. To answer this question, we study three defenses that follow this approach. Two of these are recently proposed defenses that intentionally combine components designed to work well together. A third defense combines three independent defenses. For all the components of these defenses and the combined defenses themselves, we show that an adaptive adversary can create adversarial examples successfully with low distortion. Thus, our work implies that ensemble of weak defenses is not sufficient to provide strong defense against adversarial examples.","journal":"arXiv:1706.04701 [cs]","author":[{"propositions":[],"lastnames":["He"],"firstnames":["Warren"],"suffixes":[]},{"propositions":[],"lastnames":["Wei"],"firstnames":["James"],"suffixes":[]},{"propositions":[],"lastnames":["Chen"],"firstnames":["Xinyun"],"suffixes":[]},{"propositions":[],"lastnames":["Carlini"],"firstnames":["Nicholas"],"suffixes":[]},{"propositions":[],"lastnames":["Song"],"firstnames":["Dawn"],"suffixes":[]}],"month":"June","year":"2017","note":"arXiv: 1706.04701","keywords":"⛔ No DOI found","bibtex":"@article{he_adversarial_2017,\n\ttitle = {Adversarial {Example} {Defenses}: {Ensembles} of {Weak} {Defenses} are not {Strong}},\n\tshorttitle = {Adversarial {Example} {Defenses}},\n\turl = {http://arxiv.org/abs/1706.04701},\n\tabstract = {Ongoing research has proposed several methods to defend neural networks against adversarial examples, many of which researchers have shown to be ineffective. We ask whether a strong defense can be created by combining multiple (possibly weak) defenses. To answer this question, we study three defenses that follow this approach. Two of these are recently proposed defenses that intentionally combine components designed to work well together. A third defense combines three independent defenses. For all the components of these defenses and the combined defenses themselves, we show that an adaptive adversary can create adversarial examples successfully with low distortion. Thus, our work implies that ensemble of weak defenses is not sufficient to provide strong defense against adversarial examples.},\n\tjournal = {arXiv:1706.04701 [cs]},\n\tauthor = {He, Warren and Wei, James and Chen, Xinyun and Carlini, Nicholas and Song, Dawn},\n\tmonth = jun,\n\tyear = {2017},\n\tnote = {arXiv: 1706.04701},\n\tkeywords = {⛔ No DOI found}\n}\n\n","author_short":["He, W.","Wei, J.","Chen, X.","Carlini, N.","Song, D."],"key":"he_adversarial_2017","id":"he_adversarial_2017","bibbaseid":"he-wei-chen-carlini-song-adversarialexampledefensesensemblesofweakdefensesarenotstrong-2017","role":"author","urls":{"Paper":"http://arxiv.org/abs/1706.04701"},"keyword":["⛔ No DOI found"],"downloads":0},"bibtype":"article","biburl":"https://bibbase.org/zotero/ky25","creationDate":"2019-05-11T17:47:04.243Z","downloads":0,"keywords":["⛔ no doi found"],"search_terms":["adversarial","example","defenses","ensembles","weak","defenses","strong","he","wei","chen","carlini","song"],"title":"Adversarial Example Defenses: Ensembles of Weak Defenses are not Strong","year":2017,"dataSources":["XxiQtwZYfozhQmvGR"]}