BriefGPT.xyz
Jun, 2024
揭秘神经网络的对抗脆弱性之路
Towards unlocking the mystery of adversarial fragility of neural networks
HTML
PDF
Jingchao Gao, Raghu Mudumbai, Xiaodong Wu, Jirong Yi, Catherine Xu...
TL;DR
我们研究了深度神经网络在分类任务中的对抗性鲁棒性，通过矩阵理论解释了深度神经网络对分类的对抗性脆弱性，理论结果表明输入维度增加时，神经网络的对抗性鲁棒性会降低，并且其鲁棒性只能达到最佳鲁棒性的1/√d。这一矩阵理论解释与以前的信息理论基于特征压缩的解释相一致。
Abstract
In this paper, we study the
adversarial robustness
of
deep neural networks
for
classification tasks
. We look at the smallest magnitude of
→