Code Vulnerability Detection Based on Deep Sequence and Graph Models: A Survey

With the flourishing of open-source software community, problem vulnerabilities is becoming more and serious. Hence, it urgent to come up with an effective efficient code vulnerability detection method. Source techniques used in practice today like symbolic execution fuzz testing suffer from high false positives low coverage, respectively. Traditional machine-learning-based solutions fail cope diversity vulnerabilities. To overcome these drawbacks, a large number deep-learning-based works have emerged, aiming at building powerful neural network models fully learn semantics patterns. In this survey, we mainly focus on approaches based deep sequence modeling graph technologies. Our goal investigate how two methods are applied facilitate detection. We also go over current prevailing datasets that evaluate models. At last, identify challenges field share our views future work.