The Application of Neural Models for Software

The thesis proposes methods which apply deep artificial neural networks for detecting software vulnerabilities in the scenarios where there are insufficient labeled vulnerability data available. The thesis investigates different approaches by utilizing the novel features of neural networks and the latest techniques from the field of machine learning to compensate for the lack of labeled data and facilitate the effective learning of latent vulnerable code patterns. Using the proposed methods, software code auditors can quickly pinpoint the exact location of potentially vulnerable functions, which helps to improve the accuracy and efficiency of code inspection.