Institution profile

University of the Ryukyus

Academic institutionasia · jp
Official website
Research library2linked papers
Opportunities0open roles
Selected work

Representative Papers

Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness

Aug 10, 2026

This work addresses the performance degradation of large language models (LLMs) in real-world deployment, where system prompts or safety constraints often divert generation from standard pathways despite strong benchmark results. To diagnose such robustness gaps, the authors propose a prompt-free stress-testing methodology that dynamically intervenes in the logit space during decoding by masking dominant candidate tokens at word boundaries, thereby compelling the model to generate via alternative, non-canonical paths. This approach introduces a novel runtime dynamic token-masking mechanism, establishing a zero-prompt stress-testing paradigm. Experiments across multiple open-source LLMs reveal significant effects of model scale and instruction alignment on robustness under such perturbations. Furthermore, the method efficiently produces diverse synthetic data, enabling pre-deployment reliability auditing and evaluation of safety mechanisms.

0 citationsRead paper

Contributions to conjectures in planar graphs: Induced Substructures, Treewidth, and Dominating Sets

Jun 12, 2025

This paper addresses two longstanding open conjectures on planar graphs: the Albertson–Berman Conjecture (AB), asserting that every $n$-vertex planar graph contains an induced forest of order at least $n/2$, and the Matheson–Tarjan Conjecture (MT), stating that large $n$-vertex triangulations admit a dominating set of size at most $n/4$. We develop a unified analytical framework leveraging combinatorial graph theory, structural induction, treewidth parameterization, and planar embedding analysis. Our contributions include: (i) establishing novel logical connections between connected dominating sets, induced outerplanar subgraphs, and both conjectures; (ii) constructing critical counterexamples refuting several natural generalizations; (iii) determining optimal bounds on induced subgraph sizes under various structural constraints; (iv) proving that every graph $G$ admits an induced subgraph of order at most $mathrm{tw}(G)+1$, yielding a new upper-bound tool for planar graphs; and (v) deriving tight bounds for induced linear forests and outerplanar subgraphs—advancing the resolution of both conjectures substantially.

0 citationsRead paper
Recent publications

Latest Papers

Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness

Aug 10, 2026

This work addresses the performance degradation of large language models (LLMs) in real-world deployment, where system prompts or safety constraints often divert generation from standard pathways despite strong benchmark results. To diagnose such robustness gaps, the authors propose a prompt-free stress-testing methodology that dynamically intervenes in the logit space during decoding by masking dominant candidate tokens at word boundaries, thereby compelling the model to generate via alternative, non-canonical paths. This approach introduces a novel runtime dynamic token-masking mechanism, establishing a zero-prompt stress-testing paradigm. Experiments across multiple open-source LLMs reveal significant effects of model scale and instruction alignment on robustness under such perturbations. Furthermore, the method efficiently produces diverse synthetic data, enabling pre-deployment reliability auditing and evaluation of safety mechanisms.

0 citationsRead paper

Contributions to conjectures in planar graphs: Induced Substructures, Treewidth, and Dominating Sets

Jun 12, 2025

This paper addresses two longstanding open conjectures on planar graphs: the Albertson–Berman Conjecture (AB), asserting that every $n$-vertex planar graph contains an induced forest of order at least $n/2$, and the Matheson–Tarjan Conjecture (MT), stating that large $n$-vertex triangulations admit a dominating set of size at most $n/4$. We develop a unified analytical framework leveraging combinatorial graph theory, structural induction, treewidth parameterization, and planar embedding analysis. Our contributions include: (i) establishing novel logical connections between connected dominating sets, induced outerplanar subgraphs, and both conjectures; (ii) constructing critical counterexamples refuting several natural generalizations; (iii) determining optimal bounds on induced subgraph sizes under various structural constraints; (iv) proving that every graph $G$ admits an induced subgraph of order at most $mathrm{tw}(G)+1$, yielding a new upper-bound tool for planar graphs; and (v) deriving tight bounds for induced linear forests and outerplanar subgraphs—advancing the resolution of both conjectures substantially.

0 citationsRead paper