Revisiting Vul-RAG: Reproducibility and Replicability of RAG-based Vulnerability Detection with Open-Weight Models
重新审视Vul-RAG:基于RAG的漏洞检测的可复现性与可复制性——使用开放权重模型
机构 * Institute for Secure Networked Systems, Esslingen University(安全网络系统研究所,埃斯林根大学) ; Institute for Intelligent Systems, Esslingen University(智能系统研究所,埃斯林根大学)
专题命中 推理与问题求解 :large language model(abstract);language model(abstract);分类 cs.AI
AI总结 本研究通过本地部署和多种开放权重模型,复现并扩展了Vul-RAG框架,发现其性能存在约0.30成对准确率的上限,且模型能力提升无法显著改善性能。
Comments Accepted at AI&CCPS 2026 workshop, co-located with the 21st International Conference on Availability, Reliability and Security (ARES 2026). This is the authors' preprint version