Selected work

Representative Papers

Finding Small Complete Subgraphs Efficiently

Aug 22, 2023International Workshop on Combinatorial Algorithms

This paper addresses the efficient enumeration of small cliques—particularly triangles and $K_ell$ for $ell geq 3$—in sparse graphs. Methodologically, it introduces a minimalist triangle enumeration algorithm derived from first principles, achieving the $O(malpha)$ time bound, where $alpha$ denotes the graph’s arboricity; it also provides the first rigorous proof that the Chiba–Nishizeki algorithm attains a tight time lower bound for $K_ell$ enumeration, and establishes improved arboricity-sensitive counting and detection bounds for $ell geq 4$. Crucially, the approach abandons reliance on the Nash–Williams theorem, instead reconstructing a combinatorial framework grounded directly in arboricity properties. Theoretically, the results are optimal and asymptotically tight. Experimentally, the triangle enumerator significantly outperforms baselines, while for $K_ell$ with $ell geq 7$, it achieves consistent speedups on high-arboricity graphs.

3 citations1 influentialRead paper

Graph Colouring Is Hard on Average for Polynomial Calculus and Nullstellensatz

Nov 06, 2023IEEE Annual Symposium on Foundations of Computer Science

This work investigates the average-case proof complexity of refuting 3-colorability for sparse random graphs—specifically, random regular graphs and Erdős–Rényi graphs—in Polynomial Calculus (PC) and Nullstellensatz (NS) proof systems. We establish the first linear-degree lower bounds for both graph families: over any field, any PC or NS refutation of 3-colorability requires degree Ω(n). Leveraging degree–size trade-off theorems, this implies strong exponential lower bounds of 2^Ω(n) on proof size. Our results break a long-standing barrier in average-case algebraic proof complexity, demonstrating inherent average-case hardness of 3-colorability in PC/NS. This provides a foundational benchmark for the algebraic proof complexity of random constraint satisfaction problems.

1 citationsRead paper

Testing the limits of past-adapted explanations by post-endpoint randomisation: anticipatory EEG as a worked case

Aug 12, 2026

This study addresses the challenge of distinguishing whether a model’s fit to data stems from genuine signal rather than overfitting or chance. It proposes a Level II-A inference framework that evaluates the sufficiency of explanations based solely on historical information by randomizing delay times after event endpoints are fixed. Innovatively reframing “past-explainable outcomes” from a default assumption into a quantifiable hypothesis, the approach integrates negative control probes, non-compensatory decision rules, post-hoc endpoint randomization, leakage-proof preprocessing, frozen-label blind comparators, subject eligibility verification, and sequential e-value methods. Applied to anticipatory EEG CNV data, the framework establishes bounds for spurious sufficiency on synthetic benchmarks—yielding an allocation-isolation threshold of 15 μV/s and a sequential e-value path threshold of 30 μV/s—thereby enabling conditionally valid rejections or calibrated affirmations.

0 citationsRead paper

Oil price shocks reveal unequal capacities for mobility adaptation

Aug 12, 2026

This study addresses the hidden inequalities in community adaptive capacity to urban decarbonization policies, which often increase mobility costs but whose differential impacts remain poorly observed. Leveraging the 2026 U.S.–Iran oil shock as a natural experiment, the authors employ a multi-level panel regression discontinuity design on 1.7 trillion points-of-interest visit records across 122,000 communities in China and the United States. Treating the oil price shock as an urban stress test, they reveal substantial heterogeneity in mobility responses: nearly three-quarters of communities reduced their travel range, with greater contraction among those initially exhibiting longer baseline travel distances and higher energy intensity. Communities heavily dependent on automobiles faced constrained adjustment—some sustained mobility only by bearing higher costs, while others remained structurally locked and unable to reorganize their travel patterns.

0 citationsRead paper

Stoicheia: Character-Level Masked Diffusion for Ancient Greek Textual Restoration, Parsing, and Metrical Scansion

Aug 07, 2026

This work addresses the multifaceted challenges of ancient Greek text processing—including textual restoration, word segmentation, accentuation, punctuation recovery, and morphosyntactic analysis—by proposing a character-level masked diffusion encoder with 405 million parameters. The model innovatively decomposes input into five independently maskable, aligned planes (letters, word/sentence boundaries, diacritics, case, and punctuation), enabling a unified architecture that handles all tasks without task-specific tokenization. Pretrained on an open corpus of 380 million words and rigorously evaluated using a ten-fold decontaminated cross-validation protocol to ensure generalizability, the model significantly outperforms state-of-the-art methods: it reduces character error rate in epigraphic restoration to 15.5% (a 9.1 percentage-point improvement), increases labeled attachment score (LAS) in dependency parsing by 12.9, boosts balanced accuracy in macron annotation by 6.0 points, and achieves 74.5% top-1 accuracy on the Ithaca benchmark.

0 citationsRead paper

Recent publications

Latest Papers

Testing the limits of past-adapted explanations by post-endpoint randomisation: anticipatory EEG as a worked case

Aug 12, 2026

This study addresses the challenge of distinguishing whether a model’s fit to data stems from genuine signal rather than overfitting or chance. It proposes a Level II-A inference framework that evaluates the sufficiency of explanations based solely on historical information by randomizing delay times after event endpoints are fixed. Innovatively reframing “past-explainable outcomes” from a default assumption into a quantifiable hypothesis, the approach integrates negative control probes, non-compensatory decision rules, post-hoc endpoint randomization, leakage-proof preprocessing, frozen-label blind comparators, subject eligibility verification, and sequential e-value methods. Applied to anticipatory EEG CNV data, the framework establishes bounds for spurious sufficiency on synthetic benchmarks—yielding an allocation-isolation threshold of 15 μV/s and a sequential e-value path threshold of 30 μV/s—thereby enabling conditionally valid rejections or calibrated affirmations.

0 citationsRead paper

Oil price shocks reveal unequal capacities for mobility adaptation

Aug 12, 2026

This study addresses the hidden inequalities in community adaptive capacity to urban decarbonization policies, which often increase mobility costs but whose differential impacts remain poorly observed. Leveraging the 2026 U.S.–Iran oil shock as a natural experiment, the authors employ a multi-level panel regression discontinuity design on 1.7 trillion points-of-interest visit records across 122,000 communities in China and the United States. Treating the oil price shock as an urban stress test, they reveal substantial heterogeneity in mobility responses: nearly three-quarters of communities reduced their travel range, with greater contraction among those initially exhibiting longer baseline travel distances and higher energy intensity. Communities heavily dependent on automobiles faced constrained adjustment—some sustained mobility only by bearing higher costs, while others remained structurally locked and unable to reorganize their travel patterns.

0 citationsRead paper

Stoicheia: Character-Level Masked Diffusion for Ancient Greek Textual Restoration, Parsing, and Metrical Scansion

Aug 07, 2026

This work addresses the multifaceted challenges of ancient Greek text processing—including textual restoration, word segmentation, accentuation, punctuation recovery, and morphosyntactic analysis—by proposing a character-level masked diffusion encoder with 405 million parameters. The model innovatively decomposes input into five independently maskable, aligned planes (letters, word/sentence boundaries, diacritics, case, and punctuation), enabling a unified architecture that handles all tasks without task-specific tokenization. Pretrained on an open corpus of 380 million words and rigorously evaluated using a ten-fold decontaminated cross-validation protocol to ensure generalizability, the model significantly outperforms state-of-the-art methods: it reduces character error rate in epigraphic restoration to 15.5% (a 9.1 percentage-point improvement), increases labeled attachment score (LAS) in dependency parsing by 12.9, boosts balanced accuracy in macron annotation by 6.0 points, and achieves 74.5% top-1 accuracy on the Ithaca benchmark.

0 citationsRead paper

CourseGraph: Finding overlaps and differences in Computer Science courses across universities

Aug 06, 2026

This study addresses the issue of curricular redundancy in cross-institutional course enrollment by proposing an automated method to assess substantive overlap between courses. The approach extracts course titles, descriptions, and learning outcomes, generates semantic embeddings using BERT, and employs a random forest classifier to determine whether significant overlap exists between pairs of courses. For the first time, this work formalizes the decision logic of course administrators into a computable model, enabling automatic alignment of courses across institutions. Evaluation on real-world data from Eindhoven University of Technology and Lund University demonstrates that the method effectively identifies overlapping courses, thereby providing reliable support for mutual credit recognition.

0 citationsRead paper

CyberBridge: Bridging the Gap Between Cybersecurity Education and Industry

Aug 05, 2026

This study addresses the misalignment between cybersecurity education and rapidly evolving industry competency demands. To bridge this gap, the authors propose CyberBridge, a framework that structures job postings into Knowledge, Skills, and Tasks (KST) statements, leverages Sentence-BERT for semantic embedding, and matches them to the most relevant occupational role profiles. The approach supports three key applications: career recommendation, labor market analysis, and curriculum planning. Crucially, it incorporates an explainability mechanism that traces matching outcomes back to specific competencies, thereby avoiding opaque, black-box decisions. Experimental results demonstrate that CyberBridge effectively achieves semantic alignment between job requirements and educational content, offering institutions an evidence-based tool for career guidance and curriculum optimization.

0 citationsRead paper