Proj CDeepFuzz Paper Reading: DLFuzz: Differential Fuzzing Testing of Deep Learning Systems

Abstract

本文: DLFuzz
方法: Adversarial inputs + Fuzzing, Differential Testing between before/after adding perturb
实验：
数据集：MNIST LeNet, ImageNet + VGG/ResNet
Competitor: DeepXplore
效果：

1. +338.59%的对抗性输⼊，,89.82% smaller perturbations，+2.86% neuron coverage，节省20.11%的时间消耗