※ Affiliations and titles are as of the end of the research activity.

Takamasa Iio

Development of a robot system to support the creation of social capital

Researcher

Takamasa Iio

Outline

Connections among individuals and the norms of reciprocity and trustworthiness that arise from them is called “Social capital.” Studies of sociology suggested that social capital contributes to the suppression of individual and social disorders such as loneliness, depression, poverty, and crime. In this research, we aim to develop an autonomous robot system that supports the creation of social capital by giving people an awareness of the connection with others in the community through interactions with robots.

Masaaki Imaizumi

Mathematical Detection of Adaptive Networks and Algorithm Development for Fast Deep Learning

Researcher

Masaaki Imaizumi

Outline

We will tackle the high-cost problem of deep learning by utilizing discoveries by the statistical and mathematical theory. It is well known that deep learning requires huge computational costs, and it is an obstacle of adjusting practical deep learning systems to the real society rapidly. To overcome the issue, we establish a new mathematical theory of adaptive networks for low-cost deep learning, and develop efficient algorithms based on the theoretical result. We promote the use of deep learning fit to the society by solving the cost problem.

Makoto Kato

Data Search Engine for Open Data

Researcher

Makoto Kato

Outline

This research aims to develop a data search engine for organizing the world’s open data and making it universally accessible and useful. The data search engine provides flexible and efficient access to open data by realizing queries on implications from data and data summarization through an index built by data analysis. We encourage open science and data-driven scientific decision making of the general public by establishing the data search engine.

Ryoichi Shinkuma

Network AI Enabling Data Importance Assessment for Realtime Delivery of Mobile and Spatial Information

Researcher

Ryoichi Shinkuma

Outline

Accidents and crimes in public spaces like in road transportation will be more serious issues in our future society. This study proposes a new concept of network AI, which assesses data importance in terms of contribution to the accuracy of spatial information through the machine-learning process. By prioritizing transmission of data based on the importance, more important data can be more successfully received by the server in realtime. This study aims to realize the future society in which realitime and accurate spatial information enables prevention and prompt action against accidents and crimes.

Mahito Sugiyama

Search and decomposition of higher-order interactions between variables

Researcher

Mahito Sugiyama

Outline

This research develops a methodology that searches and decomposes higher-order interactions between variables. This method will allow us to efficiently find associations from a large number of variables using advanced algorithms and measure the influence in a system in an information theoretic manner. The established technique will be applied to a wide range of applications such as statistical genetics and neuroscience to discover unknown associations between variables.

Katsuhito Sudoh

Evaluation Framework for Next-Generation Natural Language Generation

Researcher

Katsuhito Sudoh

Outline

This research aims the future in which we can access various kind of information easily with natural language, by the development of precise human and automatic evaluation schemes for natural language genearation and a natural language genaration optimized for the evaluation. Its main focus is a novel evaluation methodology for generated sentences that look very fluent but convey seriously incorrect information. Datasets for the language generation evaluation will be developed and opened to the public.

Ryu Takeda

Automatic Adaptation of Acoustic and Language Models in Spoken Dialogue System

Researcher

Ryu Takeda

Outline

The goal of this research is to develop a spoken dialogue system that is automatically customized to individual users. One of the two main technical themes is an automatic update of system’s models for automatic speech recognition during dialogues with the user. This enables the system to gradually recognize the user’s words, phrases and pronunciations that the system could not understand at first. The other is an efficient update of individual models by sharing data among several systems. This enables the system to understand utterances of new users or users that do not use the system frequently at an early stage after deployment.

Shohei Nobuhara

Proactive and cooperative distributed vision for 3D crowd motion understanding

Researcher

Shohei Nobuhara

Outline

This research is aimed at understanding 3D motion of multiple persons from multi-view images. The key challenge is incomplete observations due to severe occlusions by crowded targets themselves. This research proposes a proactive and cooperative distributed vision system which predicts crowd motion and controls observers so as to maintain visibility of the targets for robust motion understanding, in believing that such technology brings us a step closer to realizing a society with cooperative intelligent robots.

Koichi FUJIWARA

Development of medical support AI for improving quality of epilepsy treatment of non-epilepsy specialists

Researcher

Koichi FUJIWARA

Outline

Although there are more than one million epileptic patients in Japan, the number of epilepsy specialists certified by the Japan Epilepsy Society is only 600, and many of the epileptic patients have treatment by non-epilepsy specialists. It is essential to provide quality epilepsy treatment to the epileptic patients; however, it is difficult for to non-epilepsy specialists provide appropriate treatment of epilepsy.Since epilepsy specialist training requires much time and medical resources, quality of epilepsy treatment of non-epilepsy specialists need to be improved. The aim of this research is to develop medical support AI systems for improving quality of epilepsy treatment of non-epilepsy specialists.

Yuki Funabora

Flexible Active Textile with Three-dimensional Deformation and Power Transmission for Wearable Devices

Researcher

Yuki Funabora

Outline

In order to share and utilize the embodied knowledge on motions in individuals, an innovative wearable device having two features is researched: being able to wear like clothes and transmit the power directly to the user with contraction and expansion of itself. This research clarifies both design and control methods for a flexible active textile which enables to deform in three dimension and transmit three dimensional power in whole surface of the textile. Also a wearable suit for upper body with the flexible active textile will be developed, and used for test of teaching the motion knowledge between individuals to confirm the possibility to share the embodied knowledge on motions which is difficult to instruct.

Tomoyasu Horikawa

Development of brain decoding technology for generating language descriptions from human brain activity

Researcher

Tomoyasu Horikawa

Outline

Brain decoding has enabled us to reveal information represented in the brain, providing opportunities for understanding neural representations of the information and for developing brain decoding technologies that have broad applicability in many fields, including in communication and entertainment. This project aims to enhance the usability and applicability of brain decoding technologies by developing a brain decoding method that decodes sentences describing contents of presented stimuli from human brain activities measured by functional magnetic resonance imaging. For this purpose, I will build a general brain decoder that can produce language descriptions about stimuli irrespective of the differences of stimulus modalities (photographs, sounds, and texts), and investigate commonalities and differences of neural representations between different modalities.