Selected work

Representative Papers

The Price of Correlated Tests: How Strict Should a Model Release Gate Be?

Sep 30, 2026

This study addresses the issue that pass-all gating in machine learning model release tends to erroneously reject high-quality models while offering unclear reliability guarantees, by reformulating gate design as a test quantity optimization problem. This work proposes a two-class latent factor model that integrates one-dimensional integral computation with exact binomial bound verification procedures to quantify the cost of test correlation and balance reliability against retention rate. The analysis demonstrates that lenient gating is optimal under uniform correlation, revealing that test correlation substantially increases compliance costs and necessitates extensive testing when correlations are high. Furthermore, this paper provides a validation methodology for certifying gates using labeled data, thereby establishing a theoretical foundation for model release pipelines.

0 citationsRead paper

AI-to-AI Code Reviews of GitHub Pull Requests

Aug 21, 2026

研究构建了大规模AI-to-AI代码审查数据集，分析了不同AI编码代理间的代码审查行为及其变化趋势。

0 citationsRead paper

An Approach for Embedding-Guided Function Reuse Detection in Embedded C Software

Aug 04, 2026

This work addresses the challenge that embedded C functions often fail to reuse across platforms despite semantic similarity and passing static analysis, primarily due to hardware incompatibilities. To tackle this, the authors propose a domain-aware retrieval-augmented generation (RAG) approach that integrates function comments, call graphs, and project documentation, leveraging eight embedding models—including BGE, GraphCodeBERT, and LLaMA-3—for multi-perspective representation. Crucially, the method incorporates hardware abstraction layer dependencies, peripheral interfaces, and register constraints into reuse assessment for the first time. Four novel hardware compatibility verifiers are introduced: peripheral token overlap, parameter count consistency, call graph dependency overlap, and structural branching patterns. Experimental results demonstrate a verification accuracy of 97.5%, substantially outperforming SonarQube, which exhibits a false positive rate as high as 93.6%, thereby effectively identifying reuse failures caused by hardware mismatches.

0 citationsRead paper

Large-Scale Analysis of Discussions by CS Educators Across the Stack Exchange Network

Aug 04, 2026

This study investigates the professional concerns and their evolutionary trends among computer science educators on knowledge-sharing platforms. Drawing on nearly 80 million posts from the Stack Exchange network, with a focus on English-language content from the CS Educators community, the research integrates large-scale text mining, topic modeling, and manual annotation to construct, for the first time, an interpretable hierarchical taxonomy of discussion topics. The analysis comprehensively maps the discourse landscape of CS educators, revealing that programming and software development dominate as the most prevalent technical themes, while mathematics, pedagogy, and the humanities constitute significant non-technical areas of interest. These findings underscore the educators’ diverse intellectual engagement and the dynamically shifting emphases in their teaching practices.

0 citationsRead paper

Doc2CI: A Multi-Service Study of CI Configuration Generation Using Large Language Models

Aug 02, 2026

Automatically generating YAML configuration files that are both structurally valid and compliant with multiple continuous integration (CI) service specifications remains a significant challenge, and the capabilities of current large language models (LLMs) on this task are not well understood. This work introduces DOC2CI, the first cross-CI benchmark dataset comprising 3,363 document–YAML pairs, and systematically evaluates 14 open-source models alongside GPT-series models. A novel failure taxonomy is proposed to uncover the root causes of model discrepancies, and this study provides the first empirical evidence that document similarity and structural validity constitute distinct optimization objectives. Experiments reveal that even the largest models achieve an Exact Match rate below 3.1%; while 97% of generated outputs are syntactically parseable, only 71% conform to the target service schema. Schema-guided post-hoc repair without additional training boosts structural validity to 94%, whereas fine-tuning improves document similarity at the expense of standalone structural correctness.

0 citationsRead paper

Recent publications

Latest Papers

The Price of Correlated Tests: How Strict Should a Model Release Gate Be?

Sep 30, 2026

This study addresses the issue that pass-all gating in machine learning model release tends to erroneously reject high-quality models while offering unclear reliability guarantees, by reformulating gate design as a test quantity optimization problem. This work proposes a two-class latent factor model that integrates one-dimensional integral computation with exact binomial bound verification procedures to quantify the cost of test correlation and balance reliability against retention rate. The analysis demonstrates that lenient gating is optimal under uniform correlation, revealing that test correlation substantially increases compliance costs and necessitates extensive testing when correlations are high. Furthermore, this paper provides a validation methodology for certifying gates using labeled data, thereby establishing a theoretical foundation for model release pipelines.

0 citationsRead paper

AI-to-AI Code Reviews of GitHub Pull Requests

Aug 21, 2026

研究构建了大规模AI-to-AI代码审查数据集，分析了不同AI编码代理间的代码审查行为及其变化趋势。

0 citationsRead paper

An Approach for Embedding-Guided Function Reuse Detection in Embedded C Software

Aug 04, 2026

This work addresses the challenge that embedded C functions often fail to reuse across platforms despite semantic similarity and passing static analysis, primarily due to hardware incompatibilities. To tackle this, the authors propose a domain-aware retrieval-augmented generation (RAG) approach that integrates function comments, call graphs, and project documentation, leveraging eight embedding models—including BGE, GraphCodeBERT, and LLaMA-3—for multi-perspective representation. Crucially, the method incorporates hardware abstraction layer dependencies, peripheral interfaces, and register constraints into reuse assessment for the first time. Four novel hardware compatibility verifiers are introduced: peripheral token overlap, parameter count consistency, call graph dependency overlap, and structural branching patterns. Experimental results demonstrate a verification accuracy of 97.5%, substantially outperforming SonarQube, which exhibits a false positive rate as high as 93.6%, thereby effectively identifying reuse failures caused by hardware mismatches.

0 citationsRead paper

Large-Scale Analysis of Discussions by CS Educators Across the Stack Exchange Network

Aug 04, 2026

This study investigates the professional concerns and their evolutionary trends among computer science educators on knowledge-sharing platforms. Drawing on nearly 80 million posts from the Stack Exchange network, with a focus on English-language content from the CS Educators community, the research integrates large-scale text mining, topic modeling, and manual annotation to construct, for the first time, an interpretable hierarchical taxonomy of discussion topics. The analysis comprehensively maps the discourse landscape of CS educators, revealing that programming and software development dominate as the most prevalent technical themes, while mathematics, pedagogy, and the humanities constitute significant non-technical areas of interest. These findings underscore the educators’ diverse intellectual engagement and the dynamically shifting emphases in their teaching practices.

0 citationsRead paper

Doc2CI: A Multi-Service Study of CI Configuration Generation Using Large Language Models

Aug 02, 2026

Automatically generating YAML configuration files that are both structurally valid and compliant with multiple continuous integration (CI) service specifications remains a significant challenge, and the capabilities of current large language models (LLMs) on this task are not well understood. This work introduces DOC2CI, the first cross-CI benchmark dataset comprising 3,363 document–YAML pairs, and systematically evaluates 14 open-source models alongside GPT-series models. A novel failure taxonomy is proposed to uncover the root causes of model discrepancies, and this study provides the first empirical evidence that document similarity and structural validity constitute distinct optimization objectives. Experiments reveal that even the largest models achieve an Exact Match rate below 3.1%; while 97% of generated outputs are syntactically parseable, only 71% conform to the target service schema. Schema-guided post-hoc repair without additional training boosts structural validity to 94%, whereas fine-tuning improves document similarity at the expense of standalone structural correctness.

0 citationsRead paper