BriefGPT.xyz
Oct, 2019
标签平滑和Logit压缩: 对抗训练的替代方案？
Label Smoothing and Logit Squeezing: A Replacement for Adversarial Training?
HTML
PDF
Ali Shafahi, Amin Ghiasi, Furong Huang, Tom Goldstein
TL;DR
通过研究对抗训练提高分类器鲁棒性的机制，本研究表明这些机制可以有效地通过简单的正则化方法（包括标签平滑和对数挤压）及高斯噪声注入来模仿，并且在不使用对抗性示例的情况下，我们能够达到强大的对抗性鲁棒性--通常超过对抗性训练所能达到的水平。
Abstract
adversarial training
is one of the strongest defenses against adversarial attacks, but it requires
adversarial examples
to be generated for every mini-batch during optimization. The expense of producing these exa
→