Fine-Tuning Models for Automated Code Review Feedback

2026-05-14T08:51:52Z29c1f734d67e1549e75a4dd51e815df0ccc3eca7703f2de55d2b0b843dd54e65
AgentLensCode LlamaLLMPEFTSWE agentsSWE-CycleUIBenchKitVFC detectionagentic interpretationassured translationautomated code reviewbenchmarkscomparative binary analysiscozylarge language modelsprogram analysisprotocol-driven developmentsoftware assurancesoftware governancesoftware supply chainusability requirementsvulnerability-fixing commits

What happened

Collection of recent software-engineering and ML-for-code papers: key findings include that parameter-efficient fine-tuning (PEFT) of Code Llama substantially improves automated code-review feedback quality compared to prompt engineering; LLMs can extract usability requirements from user reviews but are prompt-sensitive; a formal "agentic interpretation" framework is proposed to make LLM-driven program analysis more evidence-driven; "cozy" enables comparative binary analysis to assure semantic equivalence between unsafe C code and translations to memory-safe languages; Protocol-Driven-Developm

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_se
Record identifier
29c1f734d67e1549e75a4dd51e815df0ccc3eca7703f2de55d2b0b843dd54e65
Enrichment time
2026-05-14T08:51:52Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.