International Joint Research

Japan－Singapore “AI”

The Japan Science and Technology Agency (JST) and the Singapore Agency for Science, Technology and Research (A*STAR), promote international research interaction and exchange among researchers for Japan‒Singapore collaborative research projects in the field of “AI.”

Counterpart

Agency for Science, Technology and Research (A*STAR)

Program Officer（PO）

SHIBATA, Tomohiro (Professor, Graduate School of Life Science and Systems Engineering, Kyushu Institute of Technology)

News

Projects

※The information is as of April 2025.

Project Title	Expressive and Empathetic Human-AI Interaction by Enhancing Multilingual, Multimodal Large Language Models
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	KAWAHARA, Tatsuya (Professor, School of Informatics, Kyoto University)
Singapore-side PI	Nancy CHEN (Group Leader, Institute for Infocomm Research, A*STAR)
Abstract	Current interactive AIs based on large-scale language models focus on accuracy and objectivity, and thus generate uniform responses to any user. The goal of this research is to create a multimodal conversational AI that generates responses based on language and culture, as well as speaker personality and emotion. The Japanese team will work on emotion recognition, empathic response generation, and implementation in a humanoid robot/agent, while the counterpart team will focus on multilingual and multicultural support. Furthermore, by analyzing the user's personality and emotions from his/her voice and facial expressions, we will generate emotional and empathetic responses accordingly. Not only verbal responses, but also responses through laughter and facial expressions will be realized. By mutually providing each other with the results of their research, we will realize a robot agent that can engage in multilingual and multimodal dialogue, and demonstrate it in Singapore, a multilingual and multicultural society.

Project Title	Robust Federated Foundation Model with Synthetic Data Generation
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	SAKUMA, Jun (Professor, School of Computing, Institute of Science Tokyo)
Singapore-side PI	Qingsong WEI (Principal Scientist, Institute of High Performance Computing, A*STAR)
Abstract	This collaborative research aims to build a framework for achieving safe and efficient learning of foundation models through federated learning, which allows data to be distributed while training. The Japanese team will address the issue of privacy in model learning and the issue of security in integrating distributed models. The partner team will develop a methodology for training of trusted foundation models through federated learning and address the issue of security and privacy protection. Through joint research by the teams from both countries, it is hoped that a framework will be developed to solve the security and privacy issues that are a concern in federated learning of the base models.

Project Title	Breaking Multimodal Correspondence: Crafting Safer and Fairer Multimodal AIGC
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	SATOH, Shin’ichi (Professor, Digital Content and Media Sciences Research Division, National Institute of Informatics)
Singapore-side PI	Joey ZHOU (Principal Scientist, Institute of High Performance Computing, A*STAR)
Abstract	This collaborative research seeks to address key challenges in Artificial Intelligence Generated Content (AIGC) specifically to enhance the security, fairness, and effectiveness of multimodal AI systems. As the first to tackle these issues, our objectives are to develop advanced privacy protection methods that preserve the utility of multimodal data, introduce novel anonymization techniques to address cross-modal privacy risks, and create frameworks to detect and mitigate biases in multimodal data. This project will involve close collaboration between researchers from both institutions. The backbone machine learning framework will be jointly developed. Singapore-side will put special emphasis on the first topic, Learning Fuzzy Cross-modal Correspondence for Privacy Preservation. On the other hand, Japan-side will especially address the second topic, Calibrating Demographic Distribution with Attribute Eraser for Fairness. The anticipated scientific outcomes include the development of a robust framework for anonymizing and integrating user-uploaded media, ensuring data privacy, and enhancing fairness in AI model training. We expect this research to contribute to the broader field of generative AI by providing new insights into how privacy and fairness can be simultaneously addressed in large-scale model training.

Project Title	Efficient and Private Large Multi-Modal Model Training and Inference over Heterogeneous Edge-Cloud Networks
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	CAO, Yang (Associate Professor, Department of Computer Science, Institute of Science Tokyo)
Singapore-side PI	Wei Yang Bryan LIM (Assistant Professor, College of Computing and Data Science, Nanyang Technological University)
Abstract	This collaborative research aims to develop efficient and privacy-preserving methods for training and inference of Large Multimodal Models (LMMs) across statistically heterogeneous edge-cloud networks. LMMs, which process data from various modalities such as text, images, and audio, face deployment challenges on resource-constrained devices due to their large size. To address this, the project proposes a hybrid edge-cloud approach that balances local processing for low-latency tasks with cloud-based support for more complex computations. This proposal consists of introducing Federated Parameter-Efficient Fine-Tuning (PEFT) methods for LMMs in heterogeneous environments, focusing on multi-domain and multi-modal learning; addressing efficient multi-tier inference through Mixture of Experts (MoE) and Retrieval-Augmented Generation (RAG) to route tasks between the edge and cloud; ensuring privacy for federated LMMs with new mechanisms like metric differential privacy and Trusted Execution Environments (TEEs); and including real-world case studies such as autonomous vehicles and smart manufacturing, highlighting the need for low latency and strong privacy in AI systems. The research will combine expertise from international collaborators, aiming to make significant contributions to the fields of AI, edge computing, and privacy-preserving technologies.

Project Title	Harnessing AI for Seismic Safety and Sustainability: Advancing AI-driven Technologies in Seismic Data Analysis, Subsurface Imaging, and Hazard Monitoring
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	NAGAO, Hiromichi (Associate Professor, Earthquake Research Institute, The University of Tokyo)
Singapore-side PI	Ping TONG (Associate Professor, School of Physical and Mathematical Sciences, Nanyang Technological University)
Abstract	This collaborative research aims to significantly improve AI-driven underground visualization and earthquake risk assessment technologies by developing and enhancing various AI tools for seismic data analysis based on close international collaboration between Japan and Singapore, and to contribute not only to the development of seismology but also to the utilization of underground energy and sustainable urban development. The Japan side will be responsible for improving AI technology for detecting P- and S-waves, which are the first seismic waves to arrive when an earthquake occurs, as well as for compiling Japanese seismic observation data, while the Singapore side will be responsible for compiling Singaporean seismic observation data and developing AI technology for detecting subsequent waves, which arrive later than P- and S-waves. Through this joint research by the teams from both countries, it is hoped that AI-based earthquake prediction technology will be improved, and that this will contribute to sustainable urban developments that are robust against both short- and long-period seismic vibrations.

Project Title	AI for Maritime Decarbonisation: Integrating Emerging Technologies to Vessel Navigation and Control
Project Duration	April 2025 - March 2027 (FY2025 - 2026)
Japan-side PI	HANAOKA, Shinya (Professor, School of Environment and Society, Institute of Science Tokyo)
Singapore-side PI	Ran YAN (Assistant Professor, School of Civil and Environmental Engineering, Nanyang Technological University)
Abstract	This collaborative research aims to develop an integrated model based on a developed algorithm using cutting-edge AI technologies to enhance the sustainability of international maritime transport by optimizing the navigation efficiency and decarbonization of both manned and unmanned vessels operations along the Singapore-Japan Green and Digital Shipping Corridor (GDSC). Specifically, the Japanese team will focus on collecting and processing meteorological data, verifying navigation algorithms and vessel fuel consumption prediction models, and developing and validating fine-grained ship voyage optimization models in dynamic environments and ship path-following control models. Meanwhile, the Singaporean team will be responsible for processing AIS data, developing and improving navigation solutions and vessel fuel consumption prediction models, as well as developing and enhancing fine-grained ship voyage optimization models in dynamic environments and ship path-following control models. Through this collaborative research, the project aims to significantly reduce the environmental impact of international shipping and contribute to the creation of a more sustainable future for the maritime industry.

Project Title	Advancing Electrified Transportation and Intelligent Systems through Physics-Informed Machine Learning in Wireless Power Transfer
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	FUJITA, Toshiyuki (Project Lecturer, Graduate School of Frontier Science, The University of Tokyo)
Singapore-side PI	Yi TANG (Associate Professor, School of Electrical and Electronic Engineering, Nanyang Technological University)
Abstract	This collaborative research aims to optimizing the design and performance of magnetic couplers, especially the core component and the control performance of a wireless power transfer system, for applications in electrified transportation and intelligent systems such as autonomous vehicles, personal mobility, robots, and drone through the integration of physics-informed machine learning with electromagnetic field and circuit theories. The collaboration will involve significant exchange of knowledge and personnel between Singapore and Japan. Graduate students will be dispatched for 1–6 month research visits to partner institutions, engaging in joint projects on AI and magnetic materials. The team will organize joint workshops and academic seminars, and researchers will also present findings at international conferences, ensuring the dissemination of results to the global academic community.

Project Title	AI-Driven Climate Resilient Cooling: Robust Reinforcement Learning for Mixed-Mode Ventilation
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	MIYATA, Shohei (Project Lecturer, Graduate School of Engineering, The University of Tokyo)
Singapore-side PI	Adrian CHONG (Associate Professor, Department of the Built Environment, National University of Singapore)
Abstract	This collaborative research aims to develop Mixed Mode Ventilation (MMV) technology that leverages artificial intelligence to maximize the benefits of natural ventilation during air conditioning operation. The Japanese team will build a simulation model that combines physical-based simulation and data-driven neural networks to achieve high accuracy and fast calculation speed for air conditioning equipment and indoor environments. Meanwhile, the Singapore team will provide an experimental environment and promote the development of reinforcement learning algorithms for MMV control, with a particular focus on domain adaptation. By combining the strengths of the research teams from both countries, it is hoped that more comfortable, energy-efficient and scalable MMV control will be achieved. At the same time, by implementing close information exchange and experimental cooperation with ASEAN countries, where the increase in demand for air conditioning is becoming a urgent issue, it is hoped that MMV-related technology will be deployed in the ASEAN region.

Project Title	Development of Large-scale Language and Multimodal Models for Dynamic and Sustainable Food Planning in East and Southeast Asia
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	YAMAKATA, Yoko (Professor, Information Technology Center, The University of Tokyo)
Singapore-side PI	Tat-Seng CHUA (Professor, School of Computing, National University of Singapore)
Abstract	This collaborative research aims to develop a large-scale language model (LLM) and a large-scale multimodal model (LMM) that will help people understand the situation of food in real-time by analyzing all kinds of data related to food collected via the web in response to the food situation that is dynamically changing due to climate change. Specifically, the Japanese team will collect and analyze local news and social media related to food, as well as satellite image data, etc., while also developing a food management app that analyzes and visualizes the environmental impact of people's eating habits based on their meal records. The Singapore team will lead the construction of LLM and LMM, which specialize in food using the data provided by the Japanese team. Through joint research by the teams from both countries, it is expected that people will be able to redesign food production and distribution plans in East and Southeast Asia more flexibly and quickly.

Project Title	Development of Resource-Efficient Foundation Models for Urban Heat Island Monitoring and Mitigation
Project Duration	April 2025 - March 2028 (FY2025 - 2027)
Japan-side PI	YOKOYA, Naoto (Associate Professor, Graduate School of Frontier Sciences, The University of Tokyo)
Singapore-side PI	Shijian LU (Associate Professor, College of Computing and Data Science, Nanyang Technological University)
Abstract	This collaborative research aims to develop innovative resource-efficient foundation models for monitoring and mitigating the urban heat island phenomenon. Specifically, the Japanese team will integrate multi-source data including satellite imagery, aerial photographs, and meteorological data to produce high-resolution temperature pattern estimates and 3D semantic reconstruction models, while the Singapore team will develop visual question answering and visual grounding techniques to interpret geospatial data and generate actionable recommendations for urban planning. Through joint research between the two teams, the project aims to integrate resource-efficient AI technologies and establish a new framework for enhancing urban climate resilience. Furthermore, it seeks to foster the growth of new tools and technologies for smart city management and environmental monitoring, thereby contributing to the realization of sustainable urban environments.