The study presents DefectHunter, a novel model for detecting software vulnerabilities. Traditional vulnerability detection methods are time-consuming, labor-intensive, and often underperform when dealing with complex datasets. The authors propose an innovative solution, DefectHunter, which uses the Conformer mechanism to capture local, position-wise features and global, content-based interactions. This method was found to outperform ten baseline methods in tests using six industrial and two complex datasets. The study also includes a case study demonstrating how DefectHunter understands the mechanisms underlying vulnerabilities.
Publication date: 28 Sep 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2309.15324
