BriefGPT.xyz
Sep, 2023
DiffDefense: 通过扩散模型抵御对抗性攻击
DiffDefense: Defending against Adversarial Attacks via Diffusion Models
HTML
PDF
Hondamunige Prasanna Silva, Lorenzo Seidenari, Alberto Del Bimbo
TL;DR
这篇论文提出了一种新的重建方法，利用扩散模型来保护机器学习分类器免受对抗攻击，而无需对分类器本身进行任何修改。
Abstract
This paper presents a novel
reconstruction method
that leverages
diffusion models
to protect
machine learning classifiers
against
→