Selected work

Representative Papers

Local Search for Fair Max-Min Diversification

Sep 28, 2026

This study addresses the max-min diversification problem subject to partition constraints, aiming to maximize the diversity of selected points while strictly satisfying cardinality limits for each color class. By integrating local search techniques with parameterized algorithm design, the proposed approach handles combinatorial optimization and fairness constraints in metric spaces. This work presents the first constant-factor approximation algorithm for this problem, achieving a running time of f(m)·poly(n) while exactly satisfying all constraints in polynomial time. It overcomes limitations of prior methods that suffered from high approximation ratios, exponential time complexity, or merely expected constraint satisfaction. By unifying parameterized polynomial-time computation with constant approximation guarantees, the framework is further generalized to accommodate arbitrary upper- and lower-bound constraints on subset colors.

0 citationsRead paper

Localized LoRA-MoE: Block-wise Low-Rank Experts With Adaptive Routing

Jul 06, 2026

This work addresses the limitations of existing parameter-efficient fine-tuning methods, such as LoRA, which rely on fixed architectures and struggle in dynamic scenarios involving task conflicts or sensor failures. The authors propose a novel low-rank mixture-of-experts architecture that integrates local spatial partitioning with context-aware dynamic routing, available in both block-level and unit-level variants. A key innovation is a decentralized unit-level gating mechanism that achieves performance approaching ideal global routing without requiring centralized coordination. By combining block-wise low-rank decomposition, dynamically sparse routing, and a matrix-grid coordinate field model, the method establishes gradient firewalls to suppress error propagation. Experiments on high-dimensional matrix simulation, tabular data transformation, and vision tasks under sensor degradation demonstrate substantial improvements over static baselines, achieving robust and efficient dynamic adaptation.

0 citationsRead paper

CVXPY 1.9: Recent Advances in Optimization Modeling Software

Jun 12, 2026

This work addresses the limitations of existing convex optimization modeling languages in expressiveness, solver efficiency, and problem coverage by extending the CVXPY system to establish a unified modeling and solving framework. The proposed approach introduces canonical conic quadratic programming forms, native N-dimensional expressions, explicit sparse variables, multi-attribute variables, quantum information–related cones and atoms, and a disciplined nonlinear programming (DNLP) mechanism. Coupled with a stacked-slicing backend that accelerates parameterized problems, this framework enables efficient automatic translation from user-friendly mathematical descriptions to solver-compatible inputs. The resulting system substantially enhances modeling flexibility and computational performance while broadening the scope of tractable problems, demonstrating particular advantages in quantum information applications and large-scale parametric optimization scenarios.

0 citationsRead paper

EVENT5Ws: A Large Dataset for Open-Domain Event Extraction from Documents

Apr 23, 2026

Existing event extraction datasets commonly suffer from limited event type coverage, domain closure, and a lack of large-scale human validation. To address these limitations, this work introduces EVENT5Ws—a large-scale, human-annotated, and statistically validated open-domain event extraction dataset. Through a systematic annotation pipeline and rigorous quality control mechanisms, EVENT5Ws achieves, for the first time, broad cross-regional coverage of diverse event types. The study also establishes strong baselines leveraging pretrained large language models. Experimental results demonstrate that EVENT5Ws substantially enhances model generalization across varied geographical contexts, offering a reliable resource and practical guidance for advancing open-domain event extraction research.

0 citationsRead paper

Democratizing Foundations of Problem-Solving with AI: A Breadth-First Search Curriculum for Middle School Students

Apr 01, 2026

This study investigates how to effectively integrate AI education into K–12 science classrooms—particularly in rural middle schools—to advance educational equity and accessibility. Grounded in the AI4K12 framework, the project introduces breadth-first search (BFS) as an entry point to AI problem-solving, employing device-free activities and interactive simulations that situate learning within authentic scientific contexts such as virus transmission and contact tracing. This approach enables students to grasp concepts of network exploration and shortest-path algorithms through experiential engagement. By incorporating formative assessment and learning analytics, the instructional design seamlessly blends AI literacy with disciplinary science content. Findings indicate significant improvements in students’ understanding of BFS and AI problem-solving strategies, while teacher feedback confirms strong alignment with science curriculum objectives and affirms the module’s effectiveness in supporting student learning outcomes.

0 citationsRead paper

Recent publications

Latest Papers

Local Search for Fair Max-Min Diversification

Sep 28, 2026

This study addresses the max-min diversification problem subject to partition constraints, aiming to maximize the diversity of selected points while strictly satisfying cardinality limits for each color class. By integrating local search techniques with parameterized algorithm design, the proposed approach handles combinatorial optimization and fairness constraints in metric spaces. This work presents the first constant-factor approximation algorithm for this problem, achieving a running time of f(m)·poly(n) while exactly satisfying all constraints in polynomial time. It overcomes limitations of prior methods that suffered from high approximation ratios, exponential time complexity, or merely expected constraint satisfaction. By unifying parameterized polynomial-time computation with constant approximation guarantees, the framework is further generalized to accommodate arbitrary upper- and lower-bound constraints on subset colors.

0 citationsRead paper

Localized LoRA-MoE: Block-wise Low-Rank Experts With Adaptive Routing

Jul 06, 2026

This work addresses the limitations of existing parameter-efficient fine-tuning methods, such as LoRA, which rely on fixed architectures and struggle in dynamic scenarios involving task conflicts or sensor failures. The authors propose a novel low-rank mixture-of-experts architecture that integrates local spatial partitioning with context-aware dynamic routing, available in both block-level and unit-level variants. A key innovation is a decentralized unit-level gating mechanism that achieves performance approaching ideal global routing without requiring centralized coordination. By combining block-wise low-rank decomposition, dynamically sparse routing, and a matrix-grid coordinate field model, the method establishes gradient firewalls to suppress error propagation. Experiments on high-dimensional matrix simulation, tabular data transformation, and vision tasks under sensor degradation demonstrate substantial improvements over static baselines, achieving robust and efficient dynamic adaptation.

0 citationsRead paper

CVXPY 1.9: Recent Advances in Optimization Modeling Software

Jun 12, 2026

This work addresses the limitations of existing convex optimization modeling languages in expressiveness, solver efficiency, and problem coverage by extending the CVXPY system to establish a unified modeling and solving framework. The proposed approach introduces canonical conic quadratic programming forms, native N-dimensional expressions, explicit sparse variables, multi-attribute variables, quantum information–related cones and atoms, and a disciplined nonlinear programming (DNLP) mechanism. Coupled with a stacked-slicing backend that accelerates parameterized problems, this framework enables efficient automatic translation from user-friendly mathematical descriptions to solver-compatible inputs. The resulting system substantially enhances modeling flexibility and computational performance while broadening the scope of tractable problems, demonstrating particular advantages in quantum information applications and large-scale parametric optimization scenarios.

0 citationsRead paper

EVENT5Ws: A Large Dataset for Open-Domain Event Extraction from Documents

Apr 23, 2026

Existing event extraction datasets commonly suffer from limited event type coverage, domain closure, and a lack of large-scale human validation. To address these limitations, this work introduces EVENT5Ws—a large-scale, human-annotated, and statistically validated open-domain event extraction dataset. Through a systematic annotation pipeline and rigorous quality control mechanisms, EVENT5Ws achieves, for the first time, broad cross-regional coverage of diverse event types. The study also establishes strong baselines leveraging pretrained large language models. Experimental results demonstrate that EVENT5Ws substantially enhances model generalization across varied geographical contexts, offering a reliable resource and practical guidance for advancing open-domain event extraction research.

0 citationsRead paper

Democratizing Foundations of Problem-Solving with AI: A Breadth-First Search Curriculum for Middle School Students

Apr 01, 2026

This study investigates how to effectively integrate AI education into K–12 science classrooms—particularly in rural middle schools—to advance educational equity and accessibility. Grounded in the AI4K12 framework, the project introduces breadth-first search (BFS) as an entry point to AI problem-solving, employing device-free activities and interactive simulations that situate learning within authentic scientific contexts such as virus transmission and contact tracing. This approach enables students to grasp concepts of network exploration and shortest-path algorithms through experiential engagement. By incorporating formative assessment and learning analytics, the instructional design seamlessly blends AI literacy with disciplinary science content. Findings indicate significant improvements in students’ understanding of BFS and AI problem-solving strategies, while teacher feedback confirms strong alignment with science curriculum objectives and affirms the module’s effectiveness in supporting student learning outcomes.

0 citationsRead paper