arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1902.01148cs.LGcs.CRstat.ML
收藏

通过随机化实现对抗鲁棒性的理论证据

Theoretical evidence for adversarial robustness through randomization

Université Paris-Dauphine, PSL Research University(巴黎第九大学(PSL研究大学));Institut LIST, CEA(法国原子能和替代能源委员会LIST研究所);Facebook AI Research(Facebook人工智能研究院);Wavestone(Wavestone公司);Kyoto University(京都大学);RIKEN Center for AIP(理化学研究所AIP中心)

Rafael Pinot, Laurent Meunier, Alexandre Araujo, Hisashi Kashima, Florian Yger, Cédric Gouy-Pailler, Jamal Atif

摘要:

本文研究了抵抗对抗攻击的鲁棒性理论。它侧重于随机化技术的家庭,包括在推理时在网络中注入噪声。这些技术在许多情况下被证明是有效的,但缺乏理论论据。我们通过对这些方法进行理论分析来缩小这一差距,从而解释为什么它们在实践中表现良好。更确切地说,我们做出了两个新的贡献。第一个将随机化率与对对抗攻击的鲁棒性联系起来。这一结果适用于一般的指数分布族,从而扩展和统一了以前的方法。第二个贡献在于设计了一个新的上限的对抗性推广间隙的随机神经网络。我们用一系列实验来支持我们的理论主张。

英文摘要:

This paper investigates the theory of robustness against adversarial attacks. It focuses on the family of randomization techniques that consist in injecting noise in the network at inference time. These techniques have proven effective in many contexts, but lack theoretical arguments. We close this gap by presenting a theoretical analysis of these approaches, hence explaining why they perform well in practice. More precisely, we make two new contributions. The first one relates the randomization rate to robustness to adversarial attacks. This result applies for the general family of exponential distributions, and thus extends and unifies the previous approaches. The second contribution consists in devising a new upper bound on the adversarial generalization gap of randomized neural networks. We support our theoretical claims with a set of experiments.

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。

↑