Research map / Adversarial robustness and security / Attacks on text and graphs
VFCionX: Bridging Large and Small Models for Robust Vulnerability-Fixing Commit Identification
Xing Cui, Jingzheng Wu, Wenxiang Ou, Tianyue Luo, Zhiyuan Li, Xiang Ling
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add commit message augmentation with linked issues and PRs (ablation) | VFCionX-MC · VFCI test set | F1-score (%): 75.30 -> 79.53 | Improved |
| add file selector (heuristics plus Qwen2.5-Coder-7B) (ablation) | VFCionX-PC · VFCI test set | F1-score (%): 55.26 -> 62.09 | Improved |
| add patch classifier and AdaBoost ensemble to message classifier (ablation) | VFCionX-MC · VFCI test set | F1-score (%): 79.53 -> 81.47 | Improved |
| add message classifier and ensemble to patch classifier (ablation) | VFCionX-PC · VFCI test set | F1-score (%): 62.09 -> 81.47 | Improved |
| use few-shot examples instead of zero-shot | GPT-4.5 · VFCI test set | F1-score (%): 58.72 -> 72.05 | Improved |
| use few-shot examples instead of zero-shot | Qwen2.5-7B · VFCI test set | F1-score (%): 38.57 -> 56.50 | Improved |
| use VFCionX instead of best SLM VulCurator | VulCurator · VFCI test set | F1-score (%): 69.59 -> 81.47 | Improved |
| use VFCionX instead of best LLM GPT-4.5 few-shot | GPT-4.5 (few-shot) · VFCI test set | F1-score (%): 72.05 -> 81.47 | Improved |
| use VFCionX instead of GPT-4.5 few-shot | GPT-4.5 (few-shot) · VFCI test set | Recall (%): 94.83 -> 87.09 | Hurt |
| increase max commit message length beyond 256 tokens (ablation) | VFCionX message classifier · VFCI test set | F1 | Hurt |
| increase max changed LOC threshold above about 10 lines (ablation) | VFCionX patch classifier · VFCI test set | F1 | Hurt |
Closest papers by meaning
- SecCodePRM: A Process Reward Model for Code Security (ICML 2026 · similarity 0.83)
- PatchBench: Measuring Collateral Damage in Activation Patching (NeurIPS 2026 · similarity 0.82)
- SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Tasks (NeurIPS 2025 · similarity 0.82)
- RustMizan: A Compilable, Contamination-Aware Benchmarking Framework for Rust Vulnerabilities (NeurIPS 2026 · similarity 0.82)
- IRIS: LLM-Assisted Static Analysis for Detecting Security Vulnerabilities (ICLR 2025 · similarity 0.82)
- A Large-Scale Multi-Source Dataset Linking Hacker Community Discourse to the CVE Vulnerability Lifecycle (NeurIPS 2026 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
