Model merging creates unified cross-domain code clone detector

Researchers have explored model merging techniques to create a unified model for cross-domain clone detection, addressing the fragmentation crisis caused by specialized deep learning detectors. Methods like parameter merging (TIES) and architecture merging (WUDI) were evaluated across various code models and benchmarks. TIES merging demonstrated strong generalization to unseen AI-generated clones, outperforming zero-shot code LLMs and multi-task training in robustness, despite WUDI achieving higher in-distribution performance. The study suggests that a shared pre-trained base is crucial for effective cross-model merging in software engineering. AI

IMPACT This research offers a practical method for developing more robust and generalizable code clone detection systems, potentially improving software development tools.

RANK_REASON Academic paper detailing a new methodology for model merging in software engineering. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• GPTCloneBench
• Hugging Face
• ScienceCast
• TIES
• UniXcoder
• WUDI

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →