Selected work

Representative Papers

Position: Quantum Program Generation Must Prioritize Validity Over Probabilistic Scaling

Jul 15, 2026

This work addresses the limitations of current general-purpose quantum circuit generation methods, which rely excessively on scaling model size and consequently produce outputs that frequently violate quantum-physical semantic constraints. As a result, the fraction of valid circuits decays exponentially with qubit count, rendering post-hoc filtering infeasible. To overcome this, the authors propose a verifier-centric generative architecture that embeds task-specific quantum information rules directly into the synthesis process. By integrating hierarchical constraints, topological masking, and symbolic proxies, the approach proactively guides generation to guarantee both mathematical correctness and physical validity of the output circuits. This paradigm transcends the confines of conventional imitation learning, demonstrating that merely enlarging model capacity cannot bridge the syntax–semantics gap, and establishes a novel, modular, and scalable framework for quantum program synthesis.

0 citationsRead paper

A Comprehensive Framework for Long-Term Resiliency Investment Planning under Extreme Weather Uncertainty for Electric Utilities

Apr 02, 2026

This study addresses the lack of long-term resilience investment planning methods in power systems that can simultaneously account for multiple objectives and uncertainties arising from extreme weather events, surging electricity demand, and aging infrastructure. To bridge this gap, the authors propose a novel four-stage framework that integrates digital twins of power grids, Monte Carlo simulations, and multi-objective optimization, systematically incorporating extreme weather modeling into investment decision-making for the first time. The work also provides a comparative evaluation of model-based and model-free approaches. Empirical results demonstrate that, under limited grid knowledge, a simple net present value ranking method outperforms computationally intensive model-based optimization techniques, offering a practical and efficient alternative for real-world planning applications.

0 citationsRead paper

Network-Level Travel Time Prediction Considering The Effects of Weather and Seasonality

Feb 22, 2026

This study addresses the influence of weather and seasonal factors on network-wide Travel Time Index (TTI) by proposing a machine learning prediction framework that explicitly integrates meteorological and seasonal features. Leveraging over 50,000 TTI observations collected over six years in Washington, D.C., the authors systematically incorporate weather and seasonal variables into predictive models and comparatively evaluate the performance of Ridge Regression, Support Vector Machines, and other methods for both short-term and long-term forecasting. The work presents the first quantitative assessment of weather- and season-induced effects on TTI at the network scale, demonstrating that Ridge Regression consistently outperforms competing models across all prediction tasks, yielding significantly improved accuracy. These findings offer a robust methodological foundation for intelligent transportation management under varying environmental conditions.

0 citationsRead paper

Documenting Deployment with Fabric: A Repository of Real-World AI Governance

Aug 18, 2025

Existing AI governance research predominantly emphasizes risk critique, lacking systematic empirical documentation of governance practices in real-world deployment contexts. Method: We introduce Fabric—the first scalable, visualizable repository of real-world AI governance cases—built through semi-structured interviews and collaborative workflow modeling to systematically capture, structure, and visualize governance mechanisms, oversight measures, and implementation constraints across 20 deployed AI applications. Contribution/Results: Fabric innovatively operationalizes AI governance practice into a searchable, comparable, and extensible visual case database. It identifies recurrent human oversight patterns (e.g., manual review, threshold-based intervention) and structural gaps (e.g., absent feedback loops, ambiguous accountability). Open-sourced, Fabric provides an empirical foundation and evolving platform for evidence-based AI governance research, policy formulation, and tool development.

0 citationsRead paper

Evaluating Intra-firm LLM Alignment Strategies in Business Contexts

May 24, 2025

This paper addresses the “perspective alignment” problem in enterprise large language model (LLM) assistants: implicit biases and value orientations embedded in their training data and fine-tuning objectives risk undermining critical thinking, amplifying algorithmic bias, and eroding organizational cultural integrity and ethical autonomy. Methodologically, it introduces— for the first time—theoretically grounded, internally oriented alignment strategies: supportive, adversarial, and pluralistic. Integrating instruction-tuning analysis, training-data bias diagnostics, normative ethical modeling, and organizational behavior theory, the study conducts multi-level empirical and normative analysis. Its contributions include: (1) identifying deep organizational risks arising from AI perspective misalignment; (2) proposing a balanced alignment trade-off framework that jointly satisfies technical feasibility and ethical legitimacy; and (3) delivering actionable strategic pathways and foundational theory for responsible AI governance in organizational contexts.

0 citationsRead paper

Recent publications

Latest Papers

Position: Quantum Program Generation Must Prioritize Validity Over Probabilistic Scaling

Jul 15, 2026

This work addresses the limitations of current general-purpose quantum circuit generation methods, which rely excessively on scaling model size and consequently produce outputs that frequently violate quantum-physical semantic constraints. As a result, the fraction of valid circuits decays exponentially with qubit count, rendering post-hoc filtering infeasible. To overcome this, the authors propose a verifier-centric generative architecture that embeds task-specific quantum information rules directly into the synthesis process. By integrating hierarchical constraints, topological masking, and symbolic proxies, the approach proactively guides generation to guarantee both mathematical correctness and physical validity of the output circuits. This paradigm transcends the confines of conventional imitation learning, demonstrating that merely enlarging model capacity cannot bridge the syntax–semantics gap, and establishes a novel, modular, and scalable framework for quantum program synthesis.

0 citationsRead paper

A Comprehensive Framework for Long-Term Resiliency Investment Planning under Extreme Weather Uncertainty for Electric Utilities

Apr 02, 2026

This study addresses the lack of long-term resilience investment planning methods in power systems that can simultaneously account for multiple objectives and uncertainties arising from extreme weather events, surging electricity demand, and aging infrastructure. To bridge this gap, the authors propose a novel four-stage framework that integrates digital twins of power grids, Monte Carlo simulations, and multi-objective optimization, systematically incorporating extreme weather modeling into investment decision-making for the first time. The work also provides a comparative evaluation of model-based and model-free approaches. Empirical results demonstrate that, under limited grid knowledge, a simple net present value ranking method outperforms computationally intensive model-based optimization techniques, offering a practical and efficient alternative for real-world planning applications.

0 citationsRead paper

Network-Level Travel Time Prediction Considering The Effects of Weather and Seasonality

Feb 22, 2026

This study addresses the influence of weather and seasonal factors on network-wide Travel Time Index (TTI) by proposing a machine learning prediction framework that explicitly integrates meteorological and seasonal features. Leveraging over 50,000 TTI observations collected over six years in Washington, D.C., the authors systematically incorporate weather and seasonal variables into predictive models and comparatively evaluate the performance of Ridge Regression, Support Vector Machines, and other methods for both short-term and long-term forecasting. The work presents the first quantitative assessment of weather- and season-induced effects on TTI at the network scale, demonstrating that Ridge Regression consistently outperforms competing models across all prediction tasks, yielding significantly improved accuracy. These findings offer a robust methodological foundation for intelligent transportation management under varying environmental conditions.

0 citationsRead paper

Documenting Deployment with Fabric: A Repository of Real-World AI Governance

Aug 18, 2025

Existing AI governance research predominantly emphasizes risk critique, lacking systematic empirical documentation of governance practices in real-world deployment contexts. Method: We introduce Fabric—the first scalable, visualizable repository of real-world AI governance cases—built through semi-structured interviews and collaborative workflow modeling to systematically capture, structure, and visualize governance mechanisms, oversight measures, and implementation constraints across 20 deployed AI applications. Contribution/Results: Fabric innovatively operationalizes AI governance practice into a searchable, comparable, and extensible visual case database. It identifies recurrent human oversight patterns (e.g., manual review, threshold-based intervention) and structural gaps (e.g., absent feedback loops, ambiguous accountability). Open-sourced, Fabric provides an empirical foundation and evolving platform for evidence-based AI governance research, policy formulation, and tool development.

0 citationsRead paper

Evaluating Intra-firm LLM Alignment Strategies in Business Contexts

May 24, 2025

This paper addresses the “perspective alignment” problem in enterprise large language model (LLM) assistants: implicit biases and value orientations embedded in their training data and fine-tuning objectives risk undermining critical thinking, amplifying algorithmic bias, and eroding organizational cultural integrity and ethical autonomy. Methodologically, it introduces— for the first time—theoretically grounded, internally oriented alignment strategies: supportive, adversarial, and pluralistic. Integrating instruction-tuning analysis, training-data bias diagnostics, normative ethical modeling, and organizational behavior theory, the study conducts multi-level empirical and normative analysis. Its contributions include: (1) identifying deep organizational risks arising from AI perspective misalignment; (2) proposing a balanced alignment trade-off framework that jointly satisfies technical feasibility and ethical legitimacy; and (3) delivering actionable strategic pathways and foundational theory for responsible AI governance in organizational contexts.

0 citationsRead paper