Selected work

Representative Papers

Quantum Software Engineering in Practice: FPGA and AI Integration for Quantum Certification

Jul 08, 2026

This work addresses the challenge of efficiently generating valid entangled states on noisy intermediate-scale quantum (NISQ) devices, which is hindered by hardware imperfections, noise, and decoherence. To tackle this, the authors propose QAccCert, a hybrid certification framework that uniquely integrates FPGA acceleration, artificial intelligence, and quantum software engineering methodologies. The framework leverages large language models to guide parameter optimization and employs violation of the CHSH inequality for entanglement verification. Experimental results demonstrate that, under ideal simulation conditions using Qiskit AerSimulator, the approach achieves 99.94% of the theoretical maximum Bell value (2√2), significantly outperforming random search in efficiency. These findings substantiate the feasibility and superiority of QAccCert for scalable and efficient certification of quantum devices.

0 citationsRead paper

Convex geometries and directed paths on three vertices

Jun 23, 2026

This study investigates under what conditions the $\overrightarrow{P_3}$-convexity and $\overrightarrow{P_3^*}$-convexity on directed graphs form a convex geometry—i.e., every convex set can be generated by its extreme points. By integrating graph theory, convexity theory, and computational complexity analysis, the work establishes the first necessary and sufficient condition for $\overrightarrow{P_3}$-convexity to yield a convex geometry and presents a corresponding polynomial-time recognition algorithm. In contrast, it proves that determining whether $\overrightarrow{P_3^*}$-convexity forms a convex geometry is coNP-complete in general; however, the problem becomes efficiently solvable when restricted to acyclic indifference digraphs, a special class of directed graphs.

0 citationsRead paper

Qiskit Code Migration with LLMs

Jun 18, 2026

This work addresses the challenges of code maintenance and limited reusability arising from the rapid iteration of quantum computing frameworks such as Qiskit, compounded by the tendency of general-purpose large language models to hallucinate due to a lack of high-quality, version-specific training data. To overcome these issues, the authors propose a retrieval-augmented generation (RAG) architecture grounded in a transfer scenario classification framework, integrating structured knowledge guidance with a version-aware retrieval strategy to enable automated and scalable cross-version code migration. Experimental results demonstrate that the proposed approach significantly improves both the accuracy and descriptive quality of migration recommendations. Under constrained retrieval settings, Gemini Flash-2.5 achieves superior performance on complex refactoring tasks, effectively mitigating API obsolescence.

0 citationsRead paper

Target-Side Paraphrase Augmentation for Sign Language Translation with Large Language Models

May 29, 2026

This study addresses the challenges of sign language translation, particularly the scarcity of parallel data and the long-tailed distribution of target vocabulary. The authors propose a novel approach that leverages the large language model GPT-4o to generate semantically preserved, target-controlled paraphrases, integrated with the Signformer pose-encoding architecture through a two-stage training strategy: pretraining on augmented corpora followed by fine-tuning on original data. To better evaluate semantic fidelity, they introduce an LLM-as-a-Judge mechanism, which reveals that conventional word-overlap metrics substantially underestimate actual improvements in translation quality. Experimental results demonstrate that the method achieves a BLEU-4 score of 10.33 (+0.77) on PHOENIX14T and exhibits strong generalization across GSL and LSA-T datasets, significantly enhancing semantic faithfulness.

0 citationsRead paper

Detecting UX smells in Visual Studio Code using LLMs

Feb 25, 2026

This study addresses the challenge of systematically identifying and analyzing user experience (UX) issues in Visual Studio Code, which developers frequently encounter but lack structured methodologies to diagnose. To bridge this gap, the work proposes the first application of large language models (LLMs) to automatically mine UX defects from GitHub user reports. By integrating expert-validated categorization criteria, the authors construct a reusable UX defect classification framework. Their analysis reveals that UX problems predominantly fall into four dimensions: informativeness, clarity, intuitiveness, and efficiency. These findings effectively uncover the most salient pain points in developer experience, offering a data-driven foundation for targeted improvements in integrated development environment (IDE) UX design.

0 citationsRead paper

Recent publications

Latest Papers

Quantum Software Engineering in Practice: FPGA and AI Integration for Quantum Certification

Jul 08, 2026

This work addresses the challenge of efficiently generating valid entangled states on noisy intermediate-scale quantum (NISQ) devices, which is hindered by hardware imperfections, noise, and decoherence. To tackle this, the authors propose QAccCert, a hybrid certification framework that uniquely integrates FPGA acceleration, artificial intelligence, and quantum software engineering methodologies. The framework leverages large language models to guide parameter optimization and employs violation of the CHSH inequality for entanglement verification. Experimental results demonstrate that, under ideal simulation conditions using Qiskit AerSimulator, the approach achieves 99.94% of the theoretical maximum Bell value (2√2), significantly outperforming random search in efficiency. These findings substantiate the feasibility and superiority of QAccCert for scalable and efficient certification of quantum devices.

0 citationsRead paper

Convex geometries and directed paths on three vertices

Jun 23, 2026

This study investigates under what conditions the $\overrightarrow{P_3}$-convexity and $\overrightarrow{P_3^*}$-convexity on directed graphs form a convex geometry—i.e., every convex set can be generated by its extreme points. By integrating graph theory, convexity theory, and computational complexity analysis, the work establishes the first necessary and sufficient condition for $\overrightarrow{P_3}$-convexity to yield a convex geometry and presents a corresponding polynomial-time recognition algorithm. In contrast, it proves that determining whether $\overrightarrow{P_3^*}$-convexity forms a convex geometry is coNP-complete in general; however, the problem becomes efficiently solvable when restricted to acyclic indifference digraphs, a special class of directed graphs.

0 citationsRead paper

Qiskit Code Migration with LLMs

Jun 18, 2026

This work addresses the challenges of code maintenance and limited reusability arising from the rapid iteration of quantum computing frameworks such as Qiskit, compounded by the tendency of general-purpose large language models to hallucinate due to a lack of high-quality, version-specific training data. To overcome these issues, the authors propose a retrieval-augmented generation (RAG) architecture grounded in a transfer scenario classification framework, integrating structured knowledge guidance with a version-aware retrieval strategy to enable automated and scalable cross-version code migration. Experimental results demonstrate that the proposed approach significantly improves both the accuracy and descriptive quality of migration recommendations. Under constrained retrieval settings, Gemini Flash-2.5 achieves superior performance on complex refactoring tasks, effectively mitigating API obsolescence.

0 citationsRead paper

Target-Side Paraphrase Augmentation for Sign Language Translation with Large Language Models

May 29, 2026

This study addresses the challenges of sign language translation, particularly the scarcity of parallel data and the long-tailed distribution of target vocabulary. The authors propose a novel approach that leverages the large language model GPT-4o to generate semantically preserved, target-controlled paraphrases, integrated with the Signformer pose-encoding architecture through a two-stage training strategy: pretraining on augmented corpora followed by fine-tuning on original data. To better evaluate semantic fidelity, they introduce an LLM-as-a-Judge mechanism, which reveals that conventional word-overlap metrics substantially underestimate actual improvements in translation quality. Experimental results demonstrate that the method achieves a BLEU-4 score of 10.33 (+0.77) on PHOENIX14T and exhibits strong generalization across GSL and LSA-T datasets, significantly enhancing semantic faithfulness.

0 citationsRead paper

Detecting UX smells in Visual Studio Code using LLMs

Feb 25, 2026

This study addresses the challenge of systematically identifying and analyzing user experience (UX) issues in Visual Studio Code, which developers frequently encounter but lack structured methodologies to diagnose. To bridge this gap, the work proposes the first application of large language models (LLMs) to automatically mine UX defects from GitHub user reports. By integrating expert-validated categorization criteria, the authors construct a reusable UX defect classification framework. Their analysis reveals that UX problems predominantly fall into four dimensions: informativeness, clarity, intuitiveness, and efficiency. These findings effectively uncover the most salient pain points in developer experience, offering a data-driven foundation for targeted improvements in integrated development environment (IDE) UX design.

0 citationsRead paper