Proj CDeepFuzz Paper Reading: CoCoFuzzing: Testing Neural Code Models with Coverage-Guided Fuzzing

Abstract

本文：CocoFuzzing
Task: test ML Models, test code processing models
方法：10 mutators to automatically generate validly and semantically preserving source code test cases + neuron coverage-based
实验：
对象：NeuralCodeSum, Code2Seq, Code2Vec
效果

1. 证明了能产生validly and semantically preserving examples
2. can test robustness, and generalizability and can enhance the neuron coverage
3. can use the adversarial examples to retrain models