BriefGPT.xyz
May, 2019
对抗性样本不是错误，它们是特征
Adversarial Examples Are Not Bugs, They Are Features
HTML
PDF
Andrew Ilyas, Shibani Santurkar, Dimitris Tsipras, Logan Engstrom, Brandon Tran...
TL;DR
本研究通过理论模型和实验数据，证明了对抗样本的普遍存在是由于数据分布中存在易碎且难以理解的非鲁棒特征的存在，进而解释了目前算法鲁棒性需要的人类专家知识与数据自身特性之间的不匹配性问题。
Abstract
adversarial examples
have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. We demonstrate that
adversarial examples
can be directly attrib
→