Publications

NetSight: Joint Spatio-Temporal Dependency Modeling for Network Traffic Prediction

Published in Proceedings of the 35th ACM International Conference on Information and Knowledge Management (CIKM), 2026

A joint spatio-temporal dependency modeling approach for network traffic prediction, accepted at CIKM 2026.

Evaluating Language Models on Cross-Language Code Functional Equivalence

Published in International Symposium on Empirical Software Engineering and Measurement (ESEM 2026), 2026

Introduces PolyHuman, the first dataset for evaluating cross-language code functional equivalence on human-written code, and shows LLMs are far less reliable on human-written code than on synthetic code.

Low-power and Computing-free Privacy Design for IoT Systems

Published in IEEE Computer Society Annual Symposium on VLSI (ISVLSI), 2024

Proposes a hardware-software co-design approach that leverages naturally occurring hardware noise from transistor degradation as a low-power, computing-free source of noise for Local Differential Privacy.

From GPU to TPU with Coding Agents: What Works, What Fails, and What It Costs?

Published in International Conference on Software Engineering, Software Engineering in Practice (ICSE-SEIP 2027) — Under Review, 2026

An empirical study of using Claude Code to migrate ten PyTorch ML research projects from NVIDIA GPUs to Google Cloud TPUs, assessing whether each migration runs, preserves numerical behavior, executes on the intended hardware, keeps the project goal, and performs acceptably. Currently under review.

Download Paper

Same Workflow, Different Representation: Effects on LLM-Based GitHub Actions Smell Detection

Published in IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER 2027) — Under Review, 2026

An empirical study of whether the representation through which a GitHub Actions workflow is shown to an LLM — raw YAML, a Construct Tree, or a Semantic Graph — changes configuration-smell detection. Currently under review.

Beyond Smell Detection: Construct-Aware Evaluation of Agentic LLM Pipelines for GitHub Actions Workflow Refactoring

Published in International Conference on Software Engineering (ICSE 2027) — Under Review, 2026

An empirical study of lightweight agentic LLM pipelines that both detect and refactor GitHub Actions configuration smells, comparing a joint detect-and-refactor agent against a decomposed detect-then-refactor pipeline. Currently under review.

Test Code Review in the Era of GitHub Actions: A Replication Study

Published in Under Double-Blind Review, 2025

The first analysis of GitHub Actions’ influence on test code review, leveraging 213,511 pull requests and 600,905 review comments from six large-scale GitHub projects.