Reproduce “Reassessing Dataset Quality and Fine-Tuning Practices for Vulnerability Detection on PrimeVul dataset” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Norbert Sándor Szolnoki, Gábor Antal
The authors evaluate standard code-encoder fine-tuning for whole-function vulnerability detection on PrimeVul across 770 runs of CodeBERT, GraphCodeBERT, and UniXCoder. They assess conventional scores alongside false-negative rates under a strict false-positive constraint, and test imbalance strategies and calibration. UniXCoder’s baseline reaches 0.665 F1 but misses 99.3% of vulnerabilities at a 0.5% false-positive cap; findings apply only to tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.