Multi-Language Vulnerability Detection
Software vulnerabilities pose significant threats to security. While numerous methods have been proposed to address this challenge, learning-based methods have emerged as state-of-the-art (SOTA) in performance. Notably, most existing solutions are tailored for vulnerability detection in a single programming language. Such an approach not only narrows their applicability but also overlooks potential insights from vulnerability patterns across languages, potentially leading to suboptimal performance.
Our project introduces a multi-language vulnerability detection framework, designed to identify vulnerabilities across various programming languages by leveraging datasets from diverse sources. By harnessing the commonalities in vulnerabilities across languages, our framework aims to enhance prediction accuracy for each language. Additionally, we integrate an incremental learning module, allowing for seamless expansion to new languages without necessitating access to historical data. This innovation significantly enhances the framework's practical adaptability.
Project Members
- Boyu Zhang
- Triet Le