Institution profile

Kanazawa University

Academic institutionasia · jp
Official website
Research library12linked papers
Opportunities0open roles
Selected work

Representative Papers

Waking Up an AI: A Quantitative Framework for Prompt-Induced Phase Transition in Large Language Models

Apr 16, 2025

This work investigates whether large language models (LLMs) exhibit human-like cognitive phase transitions under controlled prompting—specifically, whether they demonstrate intuitive conceptual integration akin to human cognition. Method: We introduce the novel concept of “Prompt-Induced Phase Transition” (PIPt) and propose a two-component framework comprising a Trigger Identification Protocol (TIP) and a Quantitative Prompting Protocol (TQP), enabling fine-grained, reproducible measurement of LLMs’ semantic fusion responses. Experiments systematically manipulate semantic distance, compare across models, and modulate affective and linguistic quality. Contribution/Results: We provide the first systematic empirical validation that current state-of-the-art LLMs show no significant response discontinuity under semantic fusion prompts—distinguishing them fundamentally from human intuitive conceptual integration. The study establishes the first quantitative analytical paradigm for assessing LLM cognitive behavior, revealing structural limitations in their conceptual manipulation capabilities.

1 citationsRead paper

A Cascade of Volterra-Operator BBP Transitions in a Correlated Wigner Matrix

Jul 11, 2026

This study investigates the influence of coupling strength on spectral edge phase transitions in Wigner matrices with row–column shared random factor correlation structures. By leveraging the Karhunen–Loève expansion, random matrix theory, and spectral analysis of compact Volterra integral operators, the authors demonstrate that the spectrum of the associated correlation matrix consists of a vanishing bulk component and a sequence of outlier eigenvalues. The key contribution lies in the discovery of multiple Baik–Ben Arous–Péché (BBP) phase transitions forming discrete critical levels, where the transition points are precisely determined by the singular values of the underlying compact Volterra integral operator. Theoretical predictions align with numerical simulations to within 1% error for the first twenty singular values, accurately capturing the hierarchical emergence of eigenvalues beyond the semicircle law edge.

0 citationsRead paper

General-Purpose Nonlinear Function Approximation via Linear Integrated Photonics

Jun 22, 2026

Current photonic hardware struggles to support general-purpose artificial intelligence computing due to the lack of scalable high-order nonlinear capabilities. This work proposes a hybrid optoelectronic architecture that leverages optical random Fourier feature mapping to transform high-dimensional nonlinear function approximation into linear operations, thereby enabling universal nonlinear processing on purely linear silicon photonic chips—without requiring complex nonlinear materials or active components. The approach combines scalability with high throughput and is experimentally validated through efficient implementations of tenth-order Legendre polynomials, special functions such as Voigt, Fermi-Dirac, and Fresnel profiles, neural network activation functions, two-dimensional nonlinear mappings, and a 10-dimensional softmax layer. This study thus demonstrates, for the first time, universal nonlinear computation using only linear photonic circuits.

0 citationsRead paper
Recent publications

Latest Papers

A Cascade of Volterra-Operator BBP Transitions in a Correlated Wigner Matrix

Jul 11, 2026

This study investigates the influence of coupling strength on spectral edge phase transitions in Wigner matrices with row–column shared random factor correlation structures. By leveraging the Karhunen–Loève expansion, random matrix theory, and spectral analysis of compact Volterra integral operators, the authors demonstrate that the spectrum of the associated correlation matrix consists of a vanishing bulk component and a sequence of outlier eigenvalues. The key contribution lies in the discovery of multiple Baik–Ben Arous–Péché (BBP) phase transitions forming discrete critical levels, where the transition points are precisely determined by the singular values of the underlying compact Volterra integral operator. Theoretical predictions align with numerical simulations to within 1% error for the first twenty singular values, accurately capturing the hierarchical emergence of eigenvalues beyond the semicircle law edge.

0 citationsRead paper

General-Purpose Nonlinear Function Approximation via Linear Integrated Photonics

Jun 22, 2026

Current photonic hardware struggles to support general-purpose artificial intelligence computing due to the lack of scalable high-order nonlinear capabilities. This work proposes a hybrid optoelectronic architecture that leverages optical random Fourier feature mapping to transform high-dimensional nonlinear function approximation into linear operations, thereby enabling universal nonlinear processing on purely linear silicon photonic chips—without requiring complex nonlinear materials or active components. The approach combines scalability with high throughput and is experimentally validated through efficient implementations of tenth-order Legendre polynomials, special functions such as Voigt, Fermi-Dirac, and Fresnel profiles, neural network activation functions, two-dimensional nonlinear mappings, and a 10-dimensional softmax layer. This study thus demonstrates, for the first time, universal nonlinear computation using only linear photonic circuits.

0 citationsRead paper

A thin and soft optical tactile sensor for highly sensitive object perception

Feb 03, 2026

This work proposes a novel ultrathin, soft optical tactile sensor that overcomes the limitations of conventional designs, which are typically bulky, rigid, and sensitive to optical misalignment due to complex light paths. By leveraging deformation-induced speckle pattern changes in a compliant silicone medium under applied force, the sensor enables high-precision force estimation and texture recognition through machine learning, without requiring lenses or high-end cameras. The minimalist architecture—comprising only a soft speckle-generating elastomer and a simple imaging setup—significantly enhances mechanical compliance and compactness, making it well-suited for integration into soft robotics and wearable devices. Experimental results demonstrate a low root-mean-square force estimation error of 40 mN and a 93.33% accuracy in classifying nine distinct textures, including mahjong tiles.

0 citationsRead paper