Abstract
本文: DeepGauge
Task: provide multi-granularity testing criteria for DL systems
Method: multi-granularity testing criteria for DL systems: 1. k-multisection Neuron Coverage 2. Neuron Boundary Coverage 3. Strong Neuron Activation Coverage 4. Top-k Neuron Coverage 5. Top-k Neuron Patterns
实验：
datasets: MNIST(LeNet-1, LeNet-4, LeNet-5), ImageNet(VGG-19. ResNet-50)
adversarial attack tech: Fast Gradient Sign Method (FGSM), Basic Iterative Method (BIM), Jacobian-based Saliency Map Attack (JSMA), Carlini/Wagner attack (CW)
2. Preliminaries
Existing criteria:
- Statement coverage 2. Data-flow coverage 3. Model-based coverage
- multi-granularity granularity CDeepFuzz DeepGauge criteriamulti-granularity granularity cdeepfuzz deepgauge multi-granularity category-consistent multi-granularity classification deepgauge criteria categorize according leetcode criteria granularity criteria pattern filter granularity referring grounding anything cdeepfuzz