Generating Realistic Vulnerabilities via Neural Code Editing: An Empirical Study

<p>Using a commonly used synthetic dataset and one real-world dataset, we investigate the potential and gaps of three state-of-the-art neural code editors (Graph2Edit, Hoppity, SequenceR) for DL-based realistic vulnerability data generation, and two state-of-the-art vulnerability detectors (Devign, ReVeal) to evaluate the effectiveness of the generated realistic vulnerability data.</p> <p>Once the users have Docker installed&nbsp;download the Docker image &quot;neural_editors_vulgen_docker.tar.xz&quot;.</p> <p>Then, check the README.md for detailed steps of reproducing the experiments.</p> <p>Besides, we also provide the simple package of the artifact &quot;neural_editors_vulgen.zip&quot;. The raw data of our experiments is also provided in this simple package.&nbsp;However, using it to reproduce the experiments&nbsp;requires the users to set up the enviroments and dependencies for all the five tools, which is not recommanded.</p>