Selected work

Representative Papers

Wheeler Bisimulations

Feb 08, 2026

This work addresses the lack of efficient minimization techniques for nondeterministic Wheeler automata, which stems from the absence of a suitable equivalence relation preserving their structural properties. We introduce the novel notion of *Wheeler bisimulation*, an equivalence relation that maintains the convex structure inherent to Wheeler automata, and establish—for the first time—a unique minimal form for Wheeler NFAs. Leveraging this relation, we design a linear-time algorithm that constructs the minimal Wheeler NFA in $O(m)$ time, substantially improving upon the $O(m \log n)$ complexity of conventional bisimulation-based approaches. Our results provide both theoretical foundations and practical algorithms for the efficient processing of Wheeler languages.

0 citationsRead paper

CurricuLLM: Designing Personalized and Workforce-Aligned Cybersecurity Curricula Using Fine-Tuned LLMs

Jan 08, 2026arXiv.org

This study addresses the high cost and slow iteration of cybersecurity curriculum design, which often results in a misalignment between graduates’ competencies and rapidly evolving industry skill demands. To bridge this gap, the authors propose CurricuLLM, a novel framework that leverages fine-tuned large language models for automated curriculum development. The approach first standardizes input data using PreprocessLM, then employs a fine-tuned BERT model (ClassifyLM) to accurately categorize course content into nine established knowledge domains. By integrating job-market weightings, the system dynamically generates personalized, labor-market-aligned curricula. An expert validation mechanism further enhances both the efficiency of curriculum development and its relevance to industry needs. CurricuLLM thus offers a scalable and customizable intelligent solution for modern cybersecurity education.

0 citationsRead paper

Scalable Domain-decomposed Monte Carlo Neutral Transport for Nuclear Fusion

Nov 06, 2025

The EIRENE Monte Carlo neutral transport solver lacks domain decomposition capability, suffering from severe single-node memory bottlenecks when handling ultra-large-scale meshes. Method: This work proposes and implements a Domain-Decomposed Monte Carlo (DDMC) algorithm—integrating scalable domain decomposition with neutral particle transport modeling for the first time—and embeds it within the open-source Eiron framework. Contribution/Results: The DDMC algorithm achieves strong scalability beyond linear scaling and large-scale weak scalability: on the Mahti supercomputer, it demonstrates weak scaling up to 16,384 cores, attaining 45% and 26% parallel efficiency for high- and low-collision regimes, respectively. It overcomes memory limitations, enabling previously infeasible large-scale fusion simulations. Compared to the native EIRENE algorithm, DDMC significantly improves parallel efficiency and extends the tractable problem size, establishing a scalable, high-performance computing paradigm for fusion boundary physics simulation.

0 citationsRead paper

Characterizing Small Circuit Classes from FAC^0 to FAC^1 via Discrete Ordinary Differential Equations

Jun 29, 2025

This paper investigates algebraic characterizations and complexity bounds for small Boolean circuit classes—specifically FAC⁰, FAC¹, FACC[2], and FNC¹. We introduce a unified framework based on discrete ordinary differential equations (ODEs), establishing the first exact correspondence between circuit computational power and linear-length ODE constraints: function linearity, bounded derivatives, and controlled growth rates. Our method yields ODE-completeness characterizations for FACC[2] (circuits with Mod₂ gates) and FNC¹ (log-depth circuits), delivering tight algebraic representations for multiple circuit classes. Crucially, we uncover the structural role of counting mechanisms in circuit computation, revealing how modular counting underpins separations among low-level complexity classes. These results provide novel analytical tools and potential pathways toward resolving longstanding open problems, such as the separation of FAC⁰ from FAC¹.

0 citationsRead paper

Fair Diversity Maximization with Few Representatives

Jun 09, 2025

This paper studies the fairness-constrained diversity maximization problem: selecting $k$ representative points from a multi-label dataset such that at most a constant number of samples is chosen from each label class, while maximizing the minimum pairwise distance. For small-scale cross-group representative selection, we propose the first three-stage framework—pre-pruning, padded decomposition, and label-aware assignment. We introduce a distance-driven pre-pruning step to improve efficiency; employ randomized padded decomposition to balance intra-group sparsity and global diversity; and design a label-aware cluster assignment mechanism to ensure fair label coverage. We theoretically establish an approximation ratio of $Omega(sqrt{log m}/m)$, significantly improving upon prior methods. Experiments on large-scale datasets demonstrate that our approach achieves state-of-the-art performance in both minimum pairwise distance and label coverage, while strictly satisfying fairness constraints.

0 citationsRead paper

Recent publications

Latest Papers

Wheeler Bisimulations

Feb 08, 2026

This work addresses the lack of efficient minimization techniques for nondeterministic Wheeler automata, which stems from the absence of a suitable equivalence relation preserving their structural properties. We introduce the novel notion of *Wheeler bisimulation*, an equivalence relation that maintains the convex structure inherent to Wheeler automata, and establish—for the first time—a unique minimal form for Wheeler NFAs. Leveraging this relation, we design a linear-time algorithm that constructs the minimal Wheeler NFA in $O(m)$ time, substantially improving upon the $O(m \log n)$ complexity of conventional bisimulation-based approaches. Our results provide both theoretical foundations and practical algorithms for the efficient processing of Wheeler languages.

0 citationsRead paper

CurricuLLM: Designing Personalized and Workforce-Aligned Cybersecurity Curricula Using Fine-Tuned LLMs

Jan 08, 2026arXiv.org

This study addresses the high cost and slow iteration of cybersecurity curriculum design, which often results in a misalignment between graduates’ competencies and rapidly evolving industry skill demands. To bridge this gap, the authors propose CurricuLLM, a novel framework that leverages fine-tuned large language models for automated curriculum development. The approach first standardizes input data using PreprocessLM, then employs a fine-tuned BERT model (ClassifyLM) to accurately categorize course content into nine established knowledge domains. By integrating job-market weightings, the system dynamically generates personalized, labor-market-aligned curricula. An expert validation mechanism further enhances both the efficiency of curriculum development and its relevance to industry needs. CurricuLLM thus offers a scalable and customizable intelligent solution for modern cybersecurity education.

0 citationsRead paper

Scalable Domain-decomposed Monte Carlo Neutral Transport for Nuclear Fusion

Nov 06, 2025

The EIRENE Monte Carlo neutral transport solver lacks domain decomposition capability, suffering from severe single-node memory bottlenecks when handling ultra-large-scale meshes. Method: This work proposes and implements a Domain-Decomposed Monte Carlo (DDMC) algorithm—integrating scalable domain decomposition with neutral particle transport modeling for the first time—and embeds it within the open-source Eiron framework. Contribution/Results: The DDMC algorithm achieves strong scalability beyond linear scaling and large-scale weak scalability: on the Mahti supercomputer, it demonstrates weak scaling up to 16,384 cores, attaining 45% and 26% parallel efficiency for high- and low-collision regimes, respectively. It overcomes memory limitations, enabling previously infeasible large-scale fusion simulations. Compared to the native EIRENE algorithm, DDMC significantly improves parallel efficiency and extends the tractable problem size, establishing a scalable, high-performance computing paradigm for fusion boundary physics simulation.

0 citationsRead paper

Characterizing Small Circuit Classes from FAC^0 to FAC^1 via Discrete Ordinary Differential Equations

Jun 29, 2025

This paper investigates algebraic characterizations and complexity bounds for small Boolean circuit classes—specifically FAC⁰, FAC¹, FACC[2], and FNC¹. We introduce a unified framework based on discrete ordinary differential equations (ODEs), establishing the first exact correspondence between circuit computational power and linear-length ODE constraints: function linearity, bounded derivatives, and controlled growth rates. Our method yields ODE-completeness characterizations for FACC[2] (circuits with Mod₂ gates) and FNC¹ (log-depth circuits), delivering tight algebraic representations for multiple circuit classes. Crucially, we uncover the structural role of counting mechanisms in circuit computation, revealing how modular counting underpins separations among low-level complexity classes. These results provide novel analytical tools and potential pathways toward resolving longstanding open problems, such as the separation of FAC⁰ from FAC¹.

0 citationsRead paper

Fair Diversity Maximization with Few Representatives

Jun 09, 2025

This paper studies the fairness-constrained diversity maximization problem: selecting $k$ representative points from a multi-label dataset such that at most a constant number of samples is chosen from each label class, while maximizing the minimum pairwise distance. For small-scale cross-group representative selection, we propose the first three-stage framework—pre-pruning, padded decomposition, and label-aware assignment. We introduce a distance-driven pre-pruning step to improve efficiency; employ randomized padded decomposition to balance intra-group sparsity and global diversity; and design a label-aware cluster assignment mechanism to ensure fair label coverage. We theoretically establish an approximation ratio of $Omega(sqrt{log m}/m)$, significantly improving upon prior methods. Experiments on large-scale datasets demonstrate that our approach achieves state-of-the-art performance in both minimum pairwise distance and label coverage, while strictly satisfying fairness constraints.

0 citationsRead paper