Japan Agency for Medical Research and Development (AMED)
ICT Infrastructure Establishment and Implementation of Artificial Intelligence for Clinical and Medical Research
New Support for Medical Care through IT
Principal Investigator, NII: SATOH Shin'ichi (Professor, Digital Content and Media Sciences Research Division; Director, Research Center for Medical Bigdata)
In collaboration with the medical academic societies supported by AMED, NII is undertaking the construction of big data cloud platforms for medical imaging using the Science Information NETwork (SINET5) built and operated by NII, as well as developing artificial intelligence (Al) that analyzes large collections of medical images to assist doctors with diagnosis
Construction of a Big Data Cloud Platform for Medical Imaging
The medical societies collect, anonymize, and send data from hospitals and other healthcare facilities to their respective servers. The data are then sent from each association's server to the big data cloud platform for medical imaging. To transmit confidential medical imaging information over a secure network environment, the big data cloud platform leverages the features of SINET5, which connects all the prefectures in Japan with 100 Gbps ultra-high-speed lines, as well as the enhanced virtual private network 0JPN) provided by SINIT5 (Figure). The cloud makes it possible for researchers in the medical field nationwide to safely and easily use big data for medical imaging and conduct research utilizing large volumes of data, which were previously unavailable.
Development of Al in Medical Imaging Analysis Technology
NII is carrying out a large-scale project that collects, through the medical societies, medical images of over 100,000 clinical cases across Japan, and develops medical imaging analysis technology using the core Al technologies of deep learning and image recognition. To accomplish this, NII established an R&D system in joint collaboration with other researchers in the field of informatics, and set up research themes to take on the challenge of addressing each issue.
One purpose of medical image analysis is to find subtle differences between areas with suspected lesions and normal areas within the images. Our goal is to be able to contribute to preventing oversights in the field and improving work efficiency by providing support to doctors in the fields of diagnostic imaging and medical examination.
Japan Science and Technology Agency (JST)
Strategic Basic Research Programs
ERATO: Outstanding Research Leaders Strive to Generate the Seeds of Breakthrough New Technologies
HASUO Metamathematics for Systems Design Project
Research Director: HASUO lchiro (Associate Professor, Information Systems Architecture Science Research Division; Director, Global Research Center for Systems Design and Mathematics)
In the manufacturing industry today, efforts are underway to fundamentally transform the manufacturing process--from design to production--through automation and software support using advanced information processing technologies. In light of these changes, the HASUD Metamathematics for Systems Design Project aims to introduce the findings from the field of software science into traditional manufacturing technologies and build software tools that support the different aspects of industrial product development, from developing specifications to design, implementation, and maintenance.
Leveraging Formal Methods in Manufacturing
In particular,the project explores methodologies for software support, which are responsible for quality assurance and efficiency in industrial products such as vehicles, and other "physical information systems," by bringing in the techniques of systems design in software science based on mathematics, called "formal methods. " Formal methods have so far been used with "discrete elements, "assuming calculations by computer, but in order to apply them to physical information systems, formal methods must be extended to encompass "continuous elements " of physical systems such as continuous dynamics, probability, and time (Figure). Our unique approach to this theoretically difficult problem is to mathematically analyze the process itself of extending formal methods and construct a higher-order (metalevel) theory in order to obtain universal knowledge, which may allow the various techniques of formal methods to be extended together.
This metalevel approach is very theoretical and makes full use of a variety of abstract mathematical techniques, such as logic and category theory. However, another distinctive feature of this project is its orientation towards application, with an ultimate goal of applying the outcome of these theoretical studies to the real problems faced by the industrial sector.
CREST: Network-Based(Tearn-Based) Research Giving Rise to Outstanding Results That Lead to Innovation in Science and Technology
[Big Data] Advanced Core Technologies for Big Data Integration
Research Supervisor: KITSUREGAWA Masaru (Director-General, NII)
[Research Projects] Application-Centric Overlay Cloud Using Inter-Cloud
Research Director: AIDA Kento (Professor, Information Systems Architecture Science Research Division)
With the increasing performance of supercomputers, clouds, and the networks that connect them, the possibility of building interclouds that link multiple clouds through high-performance networks and utilizing them for large-scale data processing is becoming real. However, with current technology, users have to set up computers and networks individually to build a computing platform for processing data. This poses significant technical and time barriers. The objective of this research is to develop the core technologies for quickly and automatically building large-scale data processing platforms optimized for applications on multiple clouds connected to a network. The outcome of this research will enable easy high-performance processing of large-scale data using clouds. We also aim to develop applications for genome analysis and fluid acoustic analysis in collaboration with researchers in these fields, as well as to build and operate infrastructures together with researchers at information infrastructure centers of universities and other institutions. This research is being conducted in collaboration with research groups at Hokkaido University, National Institute of Genetics, Tokyo Institute of Technology, and Kyushu University.
[Symbiotic Interaction] Creation and Development of Core Technologies Interfacing Human and Information Environments
[Research Projects] VoicePersonae: Speaker Identity Cloning and Protection
Research Director: YAMAGISHI Junichi (Professor, Digital Content and Media Sciences Research Division)
Voice is a simple, natural, and intuitive modality. At the same time, voice is also a part of our identity and is considered as an important factor in a variety of fields such as biometrics, speech synthesis, voice quality conversion, and privacy. However, research in these fields is currently being conducted separately Ioward conflicting goals. This project removes the barriers between fields related to voice identity in order to (a) refine the speaker identity modeling technology; (b) enhance the security and robustness of authentication with voice biometrics, i.e., speaker recognition; and (c) provide new technologies for voice privacy protection. Detailed modeling of speaker identity is essential for avatars that reproduce personal characteristics of individuals and other applications. Conventionally, speech synthesis, voice quality conversion, speech enhancement, and other fields were studied separately. In this project, we aim to create a new model that integrates these fields as various tasks of speech generation for multiple speakers. Aside from speech generation, we are also studying integration with speaker recognition technologies. Moreover, we will conduct research on voice biometric sensors, which are technologies for automatic detection of voice spoofing, in order to improve the safety of speaker recognition. Furthermore, we will hold the world's first challenge that will compete on voice anonymization and re-identification to accelerate research on voice privacy.
[Artificial Intelligence] Development and Integration of Artificial Intelligence Technologies for Innovation Acceleration
[Research Projects] UNDERPIN: Understanding Psychiatric Illness through Natural Language Processing and Media Analysis
Research Director: SATOH Shin'ichi (Professor, Digital Content and Media Sciences Research Division)
Psychiatric disorders such as depression, schizophrenia, and dementia are diagnosed and treated through "words." Understanding the patient's words thoroughly and bringing out the characteristic symptoms are essential to making the correct diagnosis and treatment, but objectively evaluating and quantifying them are difficult. This research aims to quantify the symptoms of psychiatric disorders using natural language processing and media analysis technologies, in order to deepen our understanding of these illnesses and eventually lead to better prevention, early detection, and other technological developments.
PRESTO: Network-Based (Individual) Research Giving Rise to Wellsprings of Innovation in Science and Technology
[Social Design] Fundamental Information Technologies towards Innovative Social System Design
Search and Decomposition of Higher-Order Interactions between Variables
Researcher: SUGIYAMA Mahito (Associate Professor』 Principles of Informatics Research Division)
Discovering and analyzing interactions between variables is a fundamental and essential challenge in analyzing multivariate data consisting of many variables. Sensors, loT, and other information technologies have been rapidly developing in recent years. This has resulted in the acquisition and collection of multivariate data on various variables over a wide range of fields, from basic sciences such as genetics, neuroscience, and social sciences to applied sciences such as medicine. Analyzing the interaction between variables in such multivariate data is one of the most basic analytical procedures in descriptive data analysis for identifying the under|ying phenomena behind the data, and is a crucial procedure in data science. To date, variable (feature) selection based on predictive analysis using linear models, such as Lasso, had been developed for machine learning, although it is not suitable for descriptive data analysis. Therefore, this research project will develop a methodology for searching and decomposing higher-order interactions between variables, as well as develop the basic theories and practical algorithms. Building discrete algorithms will enable efficient searches for higher-order interactions hidden in multivariate data consisting of many variables, as well as decomposition of these higher-order interactions using the theory of information geometry. This project's outcome will provide more sophisticated data analytics and can be applied to a wide range of fields that form the foundation of our social systems.
Cabinet Office Cross-Ministerial Strategic Innovation Promotion Program (SIP) Phase 2
Big-Data and Al-Enabled Cyberspace Technologies
This project aims to establish the cyberspace platform technologies that will florm the foundation of Society 5.0 in cyber-physical systems (CPS), in particular, (1) a highly sophisticated human interaction platform technology which will facilitate collaboration between humans and A|, (2) a cross-domain data exchange platform, and (3) inter-Al collaboration technology, with the goal of implementing CPS in the real wor|d using big data and Al.
R&D of a Cross-Domain Data Exchange Platform based on Metadata Structuring by Al Technology and its Evaluation through Applications for Spatio-temporal Big Data
Head of R&D: TAKASU Atsuhiro (Professor, Digital Content and Media Sciences Research Division)
Working together with industry partners, we are developing a cross-domain data exchange platform, a service platform for data sharing and use across different fields, and conducting studies on platform operation and maintenance. Our main action items are creating the various functions making up the platform, and providing the operational rules and support technologies to encourage its use, as well as evaluating the platform's various functions and the usefulness of the service through prototype applications. NII is particularly focused on studying information integration, which is a crilical function for linking data from different domains, and is conducting research on feature representation of linguistic data and on matching algorithms for semi-structured data contained in data from each field. We plan to demonstrate the overall usefulness of the platform by developing prototype applications that utilize data from multiple domains, and which will enable case studies that can confirm the usefulness of platform functions and identify various operational issues.