A Survey on Vulnerability Prediction using GNNs

Abstract

The massive release of software products has led to critical incidents in the software industry due to low-quality software. Software engineers lack security knowledge which causes the development of insecure software. Traditional solutions for analysing code for vulnerabilities suffer from high false positives and negative rates. Researchers over the last decade have proposed mechanisms for analysing code for vulnerabilities using machine learning. The results are promising and could replace traditional static analysis tools or accompany them in the foreseeable future to produce more reliable results. This survey presents the work done so far in vulnerability detection using Graph Neural Networks (GNNs). Presents the GNNs architectures, the graph representations, the datasets, and the results of these studies.