BriefGPT.xyz
Dec, 2014
解释和利用对抗样本
Explaining and Harnessing Adversarial Examples
HTML
PDF
Ian J. Goodfellow, Jonathon Shlens, Christian Szegedy
TL;DR
机器学习模型因神经网络的线性特性容易受到对抗性扰动的影响，该现象不同于过拟合和非线性，但可以通过生成对抗性训练样本来减小MNIST数据集中maxout网络的误差。
Abstract
Several
machine learning
models, including
neural networks
, consistently misclassify
adversarial examples
---inputs formed by applying smal
→