文章目录TaskBaselineFGSM(FastGradientSignMethod(FGSM)I-FGSM(IterativeFastGradientSignMethod)MI-FGSM(MomentumIterativeFastGradientSignMethod)M-DI2-FGSM(DiverseInputMomentumIterativeFastGradientSignMethod)ReportfgsmattackJepgCom

本文基于这篇综述进行讨论学习：《BaiT,LuoJ,ZhaoJ,etal.Recentadvancesinadversarialtrainingforadversarialrobustness[J].arXivpreprintarXiv:2102.01356,2021.》对抗训练主要目的是让模型能够对于对抗样本做出正确的判断。常见的对抗样本的生成方式（或者说