论文阅读——Imperceptible Adversarial Attack via Invertible Neural Networks
ImperceptibleAdversarialAttackviaInvertibleNeuralNetworks作者:ZihanChen,ZiyueWang,JunjieHuang*,WentaoZhao,XiaoLiu,DejianGuan解决的问题:虽然视觉不可感知性是对抗性示例的理想特性,但传统的对抗性攻击仍然会产生可追踪的对抗扰动。代码:https://github.com/jjhuan