Takuya Igaue

Internal Structure Estimation through Laser-Induced Thermal Propagation

Grant No.：JPMJAX25M1

Researcher

Takuya Igaue

Outline

This research proposes a non-contact method for estimating the internal structure of objects by observing laser-induced thermal propagation. Conventional methods, which require direct sensor contact with the concrete surface, have limited applicability for the comprehensive inspection of large-scale structures. Conversely, the high directionality of laser light enables non-contact thermal alteration of materials. Capitalizing on this property, the proposed method estimates the internal structure of concrete by analyzing the spatio-temporal dynamics of the surface temperature distribution following laser exposure.

Haruna Ishizaka

Does Generative AI Extend the Thinking of Care Professionals?

Grant No.：JPMJAX25M2

Researcher

Haruna Ishizaka

Outline

To improve the quality and efficiency of healthcare and welfare in a super-aged society, support for care professionals through AI is highly anticipated. This study focuses on the transformation of human thinking when using generative AI, aiming to promote its appropriate adoption in care settings. Through observation studies and interview surveys with care professionals, I seek to derive guidelines for designing AI that extends thinking abilities, while also clarifying directions for enabling people to collaborate with AI in a reassuring manner.

Sota Iwabuchi

Transmission Technology for Spatio-Temporal Haptic Signals

Grant No.：JPMJAX25M3

Researcher

Sota Iwabuchi

Outline

In this project, I will develop a transmission technology for spatio-temporal haptic information and, based on it, a remote haptic interaction system. As a foundation for haptic communication, I will design a new haptic stimulus data format, called UTIF, specialized for interaction. Furthermore, I will develop efficient compression and reconstruction techniques of UTIF for streaming transmission. In addition, I will build a cloud-based haptic interaction system that enables communication across heterogeneous devices, and I will conduct experimental evaluations to demonstrate its communication performance.

Yusei Onishi

Creation of Indoor Drone Platform System Using Hybrid Visible Light Sensing

Grant No.：JPMJAX25M4

Researcher

Yusei Onishi

Outline

The purpose of this research is to create fundamental technologies for indoor drones that support spatial recognition and decision-making in AI. Indoor drones function as mobile sensor nodes in AI systems, thereby contributing to the construction of cyber infrastructure for a society in which humans and AI coexist. This project develops hybrid visible light sensing technology that focuses on the reflected light from existing indoor lighting and drone-mounted lighting, and establishes positioning and three-dimensional reconstruction techniques of indoor spaces that are essential for the practical application of indoor drones.

Junichi Okamoto

Establishment of foundation for ensuring measurement data reliability in sensor networks

Grant No.：JPMJAX25M5

Researcher

Junichi Okamoto

Outline

The reliability of measurement data in sensor networks is critical for system control and operation in social infrastructure and industry. One challenge in ensuring reliability is how to guarantee that sensors are properly calibrated. This research aims to establish a foundation for ensuring reliability by rigorously describing the “specifications” that calibration data for sensors in sensor networks must satisfy as mathematical logical expressions, thereby enabling automated verification using these expressions.

Ren Ozeki

Development of Self-Adaptive Physics Hybrid AI System for Disaster Prediction in Heterogeneous Environments

Grant No.：JPMJAX25M6

Researcher

Ren Ozeki

Outline

I aim to develop a disaster prediction system deployable across heterogeneous regions and multiple hazard types, to realize Society 5.0. To overcome challenges such as data scarcity, regional heterogeneity, and the requirement for interpretability, I propose a Physics hybrid AI modeling framework. The system employs test-time adaptation techniques to achieve effective generalization without reliance on disaster occurrence data from the target region. Moreover, the hybrid model is regulated so that physics components remain the primary drivers of prediction, thereby ensuring interpretability and scientific validity. This approach enables both adaptability to heterogeneous environments and explainable disaster prediction, supporting reliable disaster management in diverse real-world contexts.

Tomoya Kitamura

Predictive Compensation Control for Haptic Transmission under Communication Delay

Grant No.：JPMJAX25M7

Researcher

Tomoya Kitamura

Outline

In teleoperation systems affected by communication delays, even slight latency or jitter can lead to operational discomfort and potential safety risks. This study proposes a predictive compensation control method that anticipates the future impedance of the target object based on visual inputs and extensive historical data. The system effectively provides a seamless tactile experience without perceptible delay by using these predictions to complement the transmitted haptic feedback. Assuming imperfect network conditions, the proposed approach aims to achieve more intuitive and reliable haptic communication, contributing to developing next-generation cyber infrastructure.

Hayato Kimura

Automated Security Evaluation Framework for Cryptographic Protocols via Cross-Layer Analysis

Grant No.：JPMJAX25M8

Researcher

Hayato Kimura

Outline

As generative AI becomes increasingly widespread, the amount of confidential data managed within Cyber Infrastructure (CI) is growing rapidly. To mitigate risks of insider misuse and configuration errors in data centers, the confidentiality and integrity of cryptographic protocols are critical. Looking ahead, the expected expansion of V2X communications and inter-operator AI agent collaboration will demand new protocols with real-time performance and high availability, alongside rigorous safety evaluations. However, current methods face two key shortcomings: they cannot jointly verify both protocol design and implementation, and they fall short in detecting vulnerabilities caused by interactions among multiple protocols. This project addresses these challenges by developing a comprehensive evaluation framework that ensures stronger and more reliable security for future CI systems.

Takaki Kiyozumi

Real-Time Reflectometry for Optical Fiber Network Damage Detection

Grant No.：JPMJAX25M9

Researcher

Takaki Kiyozumi

Outline

Optical fiber networks, the backbone of cyberinfrastructure, are vulnerable to physical attacks; as recent subsea cable breaks show, minor damage can trigger large-scale outages. This study advances Optical Correlation-Domain Reflectometry to establish a foundational technology that localizes break points in long-haul fibers with high accuracy and in real time. By enabling rapid pinpointing of damage, it shortens restoration time during incidents and strengthens the resilience of cyberinfrastructure for an AI-coexistent society.

Makoto Kobayashi

Opportunistic Sensing of Wireless Information Transmission Focusing on Electromagnetic Phenomena

Grant No.：JPMJAX25MA

Researcher

Makoto Kobayashi

Outline

Creating a truly comfortable and safe environment for humans demands not just integration, but a seamless fusion of physical space with a highly flexible cyber space. Traditional approaches relied on deploying countless wireless sensor nodes, yet such large-scale deployment has proven impractical due to severe physical and power constraints. This project takes a radically different path: it aims to realize a genuine digital twin by developing technology that senses the physical world without any dedicated sensors, instead opportunistically exploiting radio waves already used for wireless communication. By transforming ordinary communication signals into powerful sensing resources, this project overcomes the barriers of cost and scale, and pioneers a new paradigm of cyber-physical integration.

Sayako Shimizu

Design and Implementation of a Continuous Identity Infrastructure Supporting the Sustainability of Academic Careers

Grant No.：JPMJAX25MB

Researcher

Sayako Shimizu

Outline

To address the structural challenge whereby changes in researchers’ institutional affiliations result in ID alterations that hinder the continuous use of academic services, this study proposes a matching model that ensures the continuity of identity. By integrating AI-based identity relevance estimation with federated authentication, the model supports personal authenticity while also meeting institutional and ethical requirements. The research theoretically explores the co-evolution of information infrastructure, institutional design, and AI-based identity matching technologies, with the aim of inclusively accommodating diverse groups of researchers.

Yoshifumi Shu

Resource-Efficient Linux Execution Environment for Embedded AI Applications

Grant No.：JPMJAX25MC

Researcher

Yoshifumi Shu

Outline

This project aims to realize a Linux execution environment optimized for high-efficiency hardware designed for embedded AI applications by enabling the co-existence of a real-time OS and a resource-efficient Linux. This facilitates the development of feature-rich embedded AI applications while achieving superior SWaP-C (Size, Weight, Power, and Cost), real-time capability, and reliability.

Yuta Seino

Building a Dental Federated Learning Platform for Low-Resource Nodes

Grant No.：JPMJAX25MD

Researcher

Yuta Seino

Outline

Federated learning is a training methodology that does not require data aggregation, making it a promising approach in the medical field from a privacy-preserving perspective. However, the vast majority of the approximately 67,000 dental clinics nationwide are “low-resource nodes” with CPU-only environments and small datasets. Therefore, this proposal explores the potential of distributed learning across a network of dental clinics—a system characterized by a large number of asynchronous, low-resource nodes that has not been considered in previous research. My objective is to establish an AI infrastructure that can provide equitable diagnostic opportunities for all.

Kohei Taniguchi

Postal-Style Networking with Transparent Proxies

Grant No.：JPMJAX25ME

Researcher

Kohei Taniguchi

Outline

This work proposes Postal-Style Networking, which decouples network management from container management. Postal-style means packets are sent and received asynchronously, like posting to and collecting from mailboxes, and packet delivery follows policies defined by administrators. Transparent proxies terminate and aggregate container traffic, allowing containers to be paused and relocated without disrupting communication. Unified congestion control on the proxy and multipath forwarding provide fair, high throughput communication.

Hikaru Tsuchida

Secure Multi-party Computation against Multiple Adversaries

Grant No.：JPMJAX25MF

Researcher

Hikaru Tsuchida

Outline

This study aims to develop Multi-Party Computation (MPC) that demonstrates resilience against both the falsification of computational results and collusion among multiple adversaries under realistic attack scenarios. I propose a novel MPC scheme that can be constructed using only addition operations. It achieves improved efficiency and stronger security compared with existing works. The MPC scheme realized through this research is expected to make a significant contribution to the advancement and establishment of future privacy-preserving infrastructures for secure data analysis.

Yu Yamaoka

Quantum circuit model for the sense of conviction

Grant No.：JPMJAX25MG

Researcher

Yu Yamaoka

Outline

To explore the boundary where human conscious processing can be substituted by AI technologies, I’ll clarify the conditions under which the “sense of conviction” arises, based on the method and content of information presentation, by shifting information processing into the subconscious. I’ll conduct experiments that guide thought into the subconscious and measure the emergence of conviction. By comparing subjective responses about conviction (provided by participants as weak teachers) with neurophysiological data recorded during the experiments, I’ll identify human state variables associated with conviction. Each of these state variables will then be regarded as a quantum bit, and the stimulus groups as quantum circuits, in order to construct a quantum circuit model of conviction generation.

Shunpei Yamaguchi

Construction of a Visual-Inertial-Radio Odometry Platform for Unconstrained Spatial Computing

Grant No.：JPMJAX25MH

Researcher

Shunpei Yamaguchi

Outline

Based on the academic question: to what extent can virtual space remain continuously aligned with physical space?, this study aims to realize spatial computing through AR glasses that preserves user-perceived quality anytime and anywhere. To achieve this goal, this project proposes a visual-inertial-radio odometry platform that ensures highly accurate and robust self-localization, the foundation of spatial computing, thereby establishing spatial computing free from spatiotemporal constraints.

Naoki Yamaguchi

Picosecond-Pulsed CDMA LiDAR

Grant No.：JPMJAX25MI

Researcher

Naoki Yamaguchi

Outline

In this study, I employ asynchronous optical sampling to achieve picosecond pulse-based measurements, attaining a ranging resolution of 150 µm. By further introducing optical pulse coding, I realize a next-generation ToF-LiDAR. This technology enables high-definition 3D imaging and aims to accelerate advanced cyber-physical integration in social infrastructures, within an AI-symbiotic society.

Haruki Yonekura

A Trustworthy and Robust Network-Distributed MoE-LLM System

Grant No.：JPMJAX25MJ

Researcher

Haruki Yonekura

Outline

For the practical deployment of thr Internet-distributed Mixture-of-Experts (MoE) large language models, ensuring “resilience” is essential in addition to efficiency. This research addresses this critical challenge by developing and implementing: cooperative routing that considers the dynamic reliability of Experts, resistance mechanisms that suppress cascading failures and malicious Expert activity, and metrics that quantitatively evaluate system robustness. These achievements will be released as open-source software (OSS), contributing to the construction and widespread adoption of a reliable next-generation distributed AI infrastructure.

Fangzheng Lin

Efficient Detection of Spectre Gadgets with Speculation Dry Run

Grant No.：JPMJAX25MK

Researcher

Fangzheng Lin

Outline

Spectre vulnerabilities pose new threats computer system security by leaking of confidential information from programs, thus requiring urgent countermeasures. This study focuses on the acceleration and practicality enhancement of Spectre gadget detection, a key component of Spectre mitigation. I propose a novel method that integrates static analysis with Speculation Dry Run on a specialized processor architecture, addressing major performance bottlenecks of conventional approaches. My technique is expected to significantly improve efficiency while also opening the path toward real-time detection and defense, highlighting the potential for practical deployments.