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

Yutaka Arakawa

Empirical research on a information-driven social system based on a human activity recognition and intentional behavior intervention

Researcher

Yutaka Arakawa

Outline

In this research, the system that can recognize and predict a human behavior by using multiple sensors and deep learning will be developed. Based on the prediction, the intentional intervention for prompting a human behavior change will be additionally developed. Through the collaboration with actual companies, various scenario of behavior change caused by the information will be clarified empirically.

Hisashi Ishihara

Functionalization of a child android robot as a collector of deep communication data

Researcher

Hisashi Ishihara

Outline

This study aims to functionalize a child android robot so that it can actively collect deep communication data through physical and multimodal interaction with humans. The collected human’s reaction data are utilized to sophisticate each body element of the robot to induce more plentiful and deeper reactions such as unconscious and emotional ones from humans. In addition, the collected data are organized by being linked to the robot actions at the time to investigate what kinds of robot actions are effective to induce humans “obscured”reactions in social context.

Tomoyuki Kaneko

Enabling understanding and communication with intelligent AI agents

Researcher

Tomoyuki Kaneko

Outline

The aim of this research is to develop fundamental techniques to enable cooperation and mutual understanding to be actually achieved among artificial intelligence (AI) systems and human experts. Tno utilize AI systems in practice, it is essential to make intelligent AI systems appropriately explain their thoughts, so that human experts can understand and evaluate the suggestions they make. We use the games of Go and Shogi, both of which are no-luck, all-skill games, to evaluate our techniques because AI systems specializing in these domains often perform better than human experts and introduce brand-new theory, indicating forerunners of the future.

Yoshimasa Kawazoe

Toward preemptive medicine: developing a new method for analysis of large-scale electronic health records

Researcher

Yoshimasa Kawazoe

Outline

This research aims to develop a new method for early prediction of undesirable event that can occur in patients by effectively utilizing longitudinal electronic health records. By applying this method, it is expected to provide safer medical care by early predicting and dealing with events related to medical safety, such as cerebral infarction and deep venous thrombosis, which are unavoidably occurring during hospitalization.

Hiroya Takamura

Development of versatile technology for describing various types of data as text flexibly

Researcher

Hiroya Takamura

Outline

Various types of data including fincial data, physiological data, human behavioral data, have been accumulated in a large amount. I will develop a versatile technology for explaining the important parts of such data as text. This technology will bring out the potential of the accumulated data. I will also develop methods for flexibly controlling the various characteristics of generated text such as the length and the difficulty.

Yuichi Tanaka

Signal and Information Processing over Networks: Fundamental Technologies for a Better World

Researcher

Yuichi Tanaka

Outline

We pursue a world without fear of disease, disaster, and crime. In this project, I consider one of fundamental technologies for realizing a better world: Analysis of data acquired from large sensor netwoks. Specifically, I study theories of signal processing on graphs, which has been a rapidly developing research area in signal and information processing. Analysis of epidemics, estimation of starting locations of wildfire, and surveillance systems with sensor swarms are promising applications.

Tomoki Toda

Development of cooperative augmented speech production technology taking advantage of human adaptability

Researcher

Tomoki Toda

Outline

Speech communication is an essential activity in our daily life. However, there exist various barriers causing difficulties in producing speech sounds to sufficiently convey message or expressions due to physical and environmental problems or individuals’ skills. To make speech communication free from these barriers, this project aims to develop an augmented speech production technique capable of taking advantage of human adaptability by introducing a new framework to elicit cooperative actions between human and system into a statistical voice conversion technique.

Shogo Fukushima

Vocabulary Learning System Based on Memory-enhancing Effect of Emotion and Motion

Researcher

Shogo Fukushima

Outline

Research in cognitive and affective neuroscience has been revealing that moderate physiological or emotional arousal enhance long-term retention of human memory. This project aims to design systems for vocabulary learning based on this memory-enhancing effect and tries to establish a trend of vocabulary learning which intervenes effectively in human memory.

Yasuko Matsubara

Fast Mining and Forecasting of Complex Time-series Event Streams

Researcher

Yasuko Matsubara

Outline

Time-series data analysis is becoming of increasingly high importance, thanks to the decreasing cost of hardware and the increasing on-line processing capability. This research project addresses two classes of tasks: (a) Automatic mining of big data streams: we present a novel method that automatically and statistically summarizes all the big time-series events and achieves a meaningful segmentation; (b) Real-time forecasting and decision-making supports: we develop a powerful system that provides long-range forecasts and helps people make better decisions about their social activities.

Makoto Yamada

Nonlinear machine learning for scientific discovery

Researcher

Makoto Yamada

Outline

In this project, we will develop a machine learning algorithm for scientific discovery. More specifically, we will develp a nonlinear feature selection algorithm for high-dimensional big data based on sparse modeling techniques. We plan to apply the proposed algorithms for a number of applications including biomarker detection and material discovery.

Koichiro Yoshino

Spoken Dialogue System based on Incremental Language Understanding and Knowledge Acquisition

Researcher

Koichiro Yoshino

Outline

This project tries to build an autonomous knowledgebase building system for dialogue system to realize an intelligent cooperative interaction between humans and machines. We build an agent system that analyzes language structures in user utterances focused on relationships between predicates and their arguments. The system automatically learns and updates the knowledgebase that the system has inside by using the analyzed results. The system also learns the policy, mapping between the observation and the system action, to balance the user demand for the task and the system demand for knowledge acquisition to realize an intelligent cooperation with users.