This page summarizes the research topics that our laboratory is focusing on. It may serve as a reference for those considering applying to NAIST or those interested in our lab when writing essays or deciding on research themes. Please take a look!

Contents

Theme 1: Edge AI Technology for Advanced Decision-Making in Various Environments

Goal

To realize Society 5.0, the widespread integration of Cyber-Physical Systems (CPS) in various scenarios is essential. By enabling high-performance AI to run on edge devices, we aim to develop technologies that facilitate advanced decision-making support using AI, even in environments where internet or cloud connectivity is limited.

Required Technologies

  • Energy Harvesting
  • Federated Learning
  • Model Compression
  • Distributed Systems

Research Examples

  • Victor Romero, Tomokazu Matsui, Yuki Matsuda, Hirohiko Suwa, Keiichi Yasumoto: Towards Opportunistic Federated Learning Using Independent Subnetwork Training, SMARTCOMP 2024: 174-181
  • Sopicha Stirapongsasuti, Shinya Misaki, Tomokazu Matsui, Hirohiko Suwa, Keiichi Yasumoto: Batterfly: Battery-Free Daily Living Activity Recognition System through Distributed Execution over Energy Harvesting Analog PIR Sensors. DCOSS 2021: 54-56​​
  • Keiichi Yasumoto, Hirozumi Yamaguchi, Hiroshi Shigeno: Survey of Real-time Processing Technologies of IoT Data Streams. J. Inf. Process. 24(2): 195-202 (2016)​​

Theme 2: Privacy-Preserving Technologies for Secure Data Distribution

Goal

To realize Society 5.0, real-time data collection across various scenarios is essential. When handling human-related data, it is crucial to protect both objective privacy (the risk of personal identification) and subjective privacy (the feeling of embarrassment if seen by others). In this research group, we aim to develop technologies that enable the secure real-time collection of human-related data using various sensors.

Required Technologies

  • Differential Privacy
  • Estimation Methods for Privacy Protection Requirements
  • k-Anonymization and Private Information Retrieval (PIR)

Research Examples

  • Sopicha Stirapongsasuti, Francis Jerome Tiausas, Yugo Nakamura, Keiichi Yasumoto: Preserving Data Utility in Differentially Private Smart Home Data. IEEE Access 12: 56571-56581 (2024)​​
  • Lucas Maris, Yuki Matsuda, Keiichi Yasumoto: Protecting Cross-Camera Person Re-Identification Data with Image Differential Privacy. SMARTCOMP 2024: 386-391
  • Francis Tiausas, Keiichi Yasumoto, Jose Paolo Talusan, Hayato Yamana, Hirozumi Yamaguchi, Shameek Bhattacharjee, Abhishek Dubey, Sajal K. Das: HPRoP: Hierarchical Privacy-preserving Route Planning for Smart Cities. ACM Trans. Cyber Phys. Syst. 7(4): 27:1-27:25 (2023)

Theme 3: Ultra-High-Precision Positioning Technology for Multiple Mobile Nodes

Goal

We aim to develop a technology that enables ultra-high-precision positioning of multiple small mobile nodes in motion without relying on a central server, by leveraging cooperative processing among nodes.

Required Technologies

  • High-Precision Ranging Technology
  • Image Processing and Signal Processing
  • Distributed Systems

Research Examples

  • Position Estimation and Control in Constellations of Thousands to Tens of Thousands of Ultra-Small Satellites
  • Swarm Nanorobot and Drone Control

Theme 4: Generative AI for Processing Various Sensor Data

Goal

Current generative AI can understand and generate text and images, but it is not yet capable of handling sensor data. This research group aims to develop technologies for constructing generative AI that can understand and generate sensor data.

Required Technologies

  • Sensing Technology with IoT
  • Generative AI Development Techniques
  • Natural Language Processing (NLP) Technology
  • Contrastive Learning

Research Examples

  • Cross-Modal Conversion of Sensor Data, Images, Audio, and Text in Various Scenarios
  • ongoing

Theme 5: Real-Time Measurement and Sharing of Spatial Information Using a 3D Point Cloud Sensing System

Goal

By capturing space as a 3D point cloud, this technology can be utilized for spatial sharing between remote users and building virtual environments for digital twins. This research group aims to develop technologies that enable easy, real-time, and secure capture and sharing of any spatial environment.

Required Technologies

  • Spatial Capture Technology Using LiDAR and Cameras
  • Point Cloud Processing Technology at the Edge
  • Point Cloud Data Streaming Technology
  • Privacy Protection Techniques for Point Cloud Data

Research Examples

  • Tatsuya Amano, Teruhiro Mizumoto, Srikant Manas Kala, Hirozumi Yamaguchi, Tomokazu Matsui, Keiichi Yasumoto: Visual Privacy Control for Metaverse and the Beyond, IEEE Pervasive Comput. 23(1): 10-17 (2024)
  • Shigetomo Sakuma, Yuki Mishima, Tomokazu Matsui, Hirohiko Suwa, Keiichi Yasumoto, Tatsuya Amano, Hirozumi Yamaguchi: Privacy Awareness of Spatial Sharing System based on 3D Point Cloud: Insights from a User Survey, PerCom Workshops 2024: 308-313

Theme 6: Development of Human-Centric Life Support Systems

Conventional behavior recognition systems (or systems that provide services based on recognition results) have not been widely adopted due to challenges such as intrusiveness, a sense of surveillance, foreignness, reliability, and recognition accuracy. This theme aims to develop human-centric life support systems based on the principles illustrated in the following diagram.

Research Areas

・Machine Learning and Deep Learning for Behavior Recognition Logic Development

・Explainability of Machine Learning Models for Behavior Prediction

・Psychological Insights for Human-Computer Interaction (HCI)

・Front-End and Back-End Development Skills for Application Implementation

Diagram: Human-Centric Life Support System

Theme 7: Development of Stress Estimation Technology for Living with Robots

Stress management is a crucial factor in achieving well-being, yet few studies have thoroughly examined how stress emerges and is alleviated in daily life. If we can identify behavioral patterns and system interventions that influence stress susceptibility, it would enable stress-free user feedback from furniture, home appliances, and healthcare systems.

Research Areas

・Fundamental Machine Learning and Deep Learning

・Psychological Insights for Understanding Stress

・Front-End and Back-End Development Skills for Application Implementation

Theme 8: Development of Multi-Resident Behavior Recognition Technology for Household Environments

Behavior recognition in multi-resident households is highly challenging, as it requires not only recognizing actions but also identifying the individual performing them. In general household settings, the use of cameras and microphones often raises privacy concerns, and the use of wearable sensors may also be limited. This theme aims to develop a multi-resident behavior recognition system designed for household use by leveraging limited sensors, such as motion sensors, environmental sensors, and privacy-conscious sensors.

Research Areas

・Applied Machine Learning and Deep Learning

・Sensing and Sensor Development Skills

画像に alt 属性が指定されていません。ファイル名: image-4.png

Theme 9: Gamification of Daily Life

To achieve self-fulfillment through goal-setting, improving daily life behaviors is often essential. Additionally, certain actions, such as cleaning a room or washing hands after coming home, are behaviors that individuals may want to develop as habits. This theme aims to develop a gamified life support system that integrates these behaviors into daily life, helping residents move closer to their ideal selves.

Research Areas

・Skills for Game and Interface Development

・Psychological Insights

・Game Design and Development

Theme 10: Realizing Well-Being-Centered Human-Computer Interaction

Background

Digital technology has deeply permeated our lives, becoming essential for both social and economic activities. However, it has also introduced new challenges, such as digital dependency, privacy invasion, and stress caused by information overload. The current digital society is not necessarily human-centered, let alone Well-being-centered. Therefore, we aim to achieve a harmonious integration of technology and humanity, striving to create a human-centered digital society.

Technology Examples

  • Human Digital Twin Technology: Comprehensive estimation of human states by integrating biometric and environmental data, along with predicting and controlling state transitions.
  • Federated Learning Architecture: Aggregating individual models to evolve collective intelligence.
  • Adaptive Information Control Theory: Optimizing the trade-off between cognitive load and information value.

Application Examples

  • Interaction Platform for Enhancing Empathy: A smartphone-integrated system designed to foster empathy through interactive experiences.
  • Music Recommendation System Based on User Characteristics: A personalized system that suggests music tailored to individual user traits and preferences.

Theme 11: Establishing Real-Time Affective Computing

Background

Our daily lives are enriched by a wide range of emotions, such as joy, sadness, anger, and anxiety, which significantly influence decision-making and behavior. However, current information systems do not fully understand or utilize human emotional states. Although technologies for estimating emotions from facial expressions, vocal tones, and biometric signals have advanced, a technological foundation for real-time emotion recognition and utilization at both individual and societal levels has yet to be established. To create a more human-centric and empathetic technology, we aim to establish a new information society that understands and leverages emotions.

Technology Examples

  • Emotion Estimation Technology Using Multimodal Sensing Data
  • Establishing Multi-Scale Emotion Propagation Models and Control Theory from Individuals to Groups
  • Emotion-Guided Architecture for Proactive Interventions Based on Emotion State Prediction
  • Establishing a Self-Regulation Mechanism to Ensure the Ethicality and Safety of Emotion AI Interventions

Application Examples

  • Protecting Individuals from the Psychological Impact of Emotion AI: Ethical validation through step-by-step empirical experiments.
  • Development of a High-Precision Emotion Estimation AI Model with Short-Term Temporal Resolution.

Theme 12: Creation of Intelligent Information Environment Computing

Background

私たちの生活空間は、急速にデジタル化が進んでいます。家庭やオフィス、都市空間全体が、インテリジェントなシステムによってつながる「知的情報環境」が形成されつつあります。しかし現状では、個々のシステムは独立して動作し、環境全体としての知的な振る舞いや、人々への自然な支援は実現できていません。環境そのものが知性を持ち、人間と共に進化しながら調和的な知の創発を実現する、新たな環境知能パラダイムの確立に挑戦します。

Technology Examples

  • Environmental Intelligence Technology: Inferring intentions and context through ubiquitous sensor networks.
  • Spatial Synchronization Technology: Ensuring consistency between physical and cyber spaces.
  • Environment Optimization Technology: Utilizing distributed cooperative learning under privacy constraints.
  • Embedded AI Systems for Real-World Adaptation: Adapting to real-world uncertainties with AI integrated into the environment.

Application Examples

  • Unconscious Learning Environment Optimization System Utilizing Ubiquitous Sensors
  • Urban Co-Creation Platform Through Participatory Environmental Sensing
  • Research on Optimizing Information Delivery Methods Adapted to User States and Urban Spaces

Theme 13: Town Sensing Based on BLE Data

Background

Real-time pedestrian flow measurement technologies using cameras and various sensors are being developed, but they have yet to achieve full coverage of urban areas. On the other hand, methods utilizing GPS data from users exist for tracking urban pedestrian flows; however, real-time processing remains a challenge. Therefore, there is a growing need for a method that enables real-time monitoring of pedestrian flows across entire urban areas while also providing future pedestrian flow predictions.

Technology Examples

  • Circulation and Stay Detection Technology Using BLE Sensing
  • City-Wide Pedestrian Flow Estimation Using Passive MobilBLE Sensing
  • Pedestrian Flow Prediction Through Multi-Agent Simulation Using Synthetic Population Data and Data Assimilation

Application Examples

  • Navigation Based on City-Wide Pedestrian Flow Estimation and Prediction
  • Optimization of Demand-Responsive Transport Based on Pedestrian Flow Predictions
  • Policy Planning for Transportation Based on Evidence-Based Policy Making (EBPM)

Theme 14: An Acceptable and Inclusive Monitoring Society

Background

The realization of Society 5.0 requires individual monitoring, but such monitoring is often met with resistance. To address this issue, it is not enough to simply enhance hardware security; a comprehensive approach is needed that includes software for managing personal data and operational frameworks. This entire system should be regarded as a social infrastructure, and efforts should be made to digitalize it as an ICT-based solution.

Technology Examples

  • Development of a Personal Information Circulation Acceptance Model
  • Building a Personal Information Circulation Infrastructure Using PPLR and PIFoT
  • Personal Information Circulation Technology Utilizing Differential Privacy and Federated Learning
  • Persuasion and Behavior Change Techniques for Personal Information Circulation Acceptance Using LLMs

Application Examples

  • Privacy Control Considering Context
  • Circulation and Management of Personal Information During Disasters
  • Monitoring and Protection of Children and the Elderly in Emergencies

Theme 15: Framework Design for Informational Health

Background

The importance of informational health is growing in response to the attention economy, exemplified by filter bubbles and echo chambers. To achieve informational health, it is essential to support self-management of information exposure. This research aims to manage the quality and quantity of attention continuously taken by information technology, using information technology itself, ultimately fostering better informational health.

Technology Examples

  • Stance Detection Using Machine Learning and Deep Learning
  • Agent Generation Using Large Language Models (LLMs)
  • Behavior Change in Information Consumption Through IoT-Based Nudging

Application Examples

  • Implementation on News Sites and Platforms
  • Rating System for News Articles
  • Digital Health Checkup
  • Digital Diet Through Balanced Information Exposure