ISD － Schanze Toward the Next Stage of ATR




Masanobu Higashida, Director
ATR International, Information Sciences Division


The Information Sciences Division (ISD) was formed within ATR International (ATR-I), the parent company of ATR R&D Laboratories, in January 2000. The initial mission for ISD was to promote research activities under contract agreements with governmental and private organizations such as the Japan Science and Technology Corporation (JST) and Nippon Telegraph and Telephone Corporation (NTT).9 In order to stride deeper into the human science field, ISD was expanded this April through the addition of researchers from ATR Human Information Processing Research Laboratories, which scaled down this year due to closure of the project in early 2001, as scheduled in the planning stage in 1993. The ISD staff presently consists of approximately forty researchers and supporting staff.

In the 14-year history of the ATR Group, the Japan Key Technology Center (KTC), which was established in 1985, has been the main provider of financial support for R&D companies. Recently, however, certain problems have become revealed with investment-type research funding. ATR-I, as the headquarters of the ATR Group, is therefore continually seeking other frameworks of support to maintain the excellent working environment that has been cultivated by our participating members, while the existing laboratories continue to be supported by KTC funds. Obtaining research funding from diverse sources under contract agreements is a part of this effort. Until a new research framework is established, ATR-I has decided to keep many distinguished scientists and engineers within the present ATR environment.

For ISD to be recognized as a cradle for basic research, ATR-I will continue to supervise ISD activities by feeding the projects with sufficient funds obtained from inside and outside of the company.

As shown in the attached figure, ISD conducts three different types of research for original purposes.

The following is a brief introduction to ISD.


1. Joint research with other research facilities

As one example, research is being conducted jointly with the Communication Research Laboratories (CRL), which is operated by the Ministry of Posts and Telecommunications. A CRL branch research center was also established this April in the ATR building to provide a synergetic effect in related research activities.

The main goal of this type of research is to propose new means of communications, to enhance the functionality of human/computer interaction for application to tomorrow's cyber-society. Plans call for research to be conducted on basic human communication functions to develop computational models that will simulate these functions. Implementation of the models and evaluations will also be examined.


2. In-house investment type-research project

One of the objectives of ATR-I is to become a center of excellence in the neuroscience field. ATR-I has provided brain-activity researchers with an effective investigation tool called fMRI (functional Magnetic Resonance Imaging). Operation and consulting services are also available to those outside of the ATR group.


3. Research funded by governmental organizations

This type of research is being supported by research foundations such as the Science and Technology Agency (STA) and the Japan Science and Technology Corporation (JST), To apply for funding, research topics are individually planned and submitted to the above organizations as multi-year research projects. The following four projects are currently underway.

1 Joint research with other research facilities
(1)  Biological Speech Science Project (PL: K.Honda)
Study of biological functions involved in speech communication Speech is the easiest and most natural means of human communication and is enabled by innate biological functions. Science and technology need to learn such capabilities to develop effective tools for human communication. To do so, we established a research project to study the biological aspects of speech, such as characteristics that are different across speakers and universal among all of us, based on the morphological analysis and functional modeling of speech production systems.
(2) Spoken Language Acquisition Project (PL: R.Yamada)
Human communication relies heavily on spoken language; we talk with others and listen to others every day, without much effort. Yet we are often not aware of the psychological mechanisms that underlie this ability. To better understand how humans process spoken language, our research group is studying how language learners develop listening and pronunciation skills in a foreign language, using state-of-the-art technology. Our research findings not only advance our understanding of human speech processing, but they also help develop better technology for human-computer interaction and effective instructional tools for second-language education.
(3) Communication Dynamics Project (PL: E.V-Bateson)
The main goal of this project is to understand how humans process communicative events in complex environments and to represent them computationally. A major focus is the analysis and synthesis of multimodal speech behavior in emotional and multi-talker environments. However, expressions of emotion and other gestural forms of communication are also being examined, as well as basic visual processes such as the detection and representation of three-dimensional objects in interactive and changing environments.
(4) Emergent Communication Mechanism Project (PL: K.Shimohara)
Postulating communications as "forms of relationships with others", we are aiming to create emergent communication mechanisms not only to activate cyber-social interactions between people and information but also to enable people to find diverse relationships. Based on evolutionary systems now being worked on, the following research topics are being pursued: (1) artificial brains and emotion, (2) organizational and socio-dynamics, (3) cyber communications, and (4) "KANSET" communications.
(5) Cyberhuman Project (PL: M.Kawato)
We investigate information processing by the human brain by making a humanoid robot solve the same computational problems as those for the brain. The main research topics are as follows.
1) Exploring how to create complex behaviors with our existing robots.
2) Research on adapting shape and texture models both in the domain of estimating and tracking of 3-D body motion as well as the case of more simple 2-D EM-based hand and object tracking.
3) Investigating reinforcement learning and learning by observation algorithms in various domains.
4) Making progress on coaching behaviors for a humanoid.
5) Exploring the mechanisms by which human movement is visually perceived and its different properties recognized.
6) Investigating the production of human-like movement through the design and analysis of interactions with the SARCOS dexterous figure.
7) Several interesting computational models of the cerebellum.
8) Simulation of perceiving biological motions with more detailed arm modeling.
2 In-house investment type-research project
(6) Introduction of Brain Activity Imaging Center Project (PL: S.Masaki)
The Main purpose of the ATR Brain Activity Imaging Center (BAIC) Project is to maintain an environment for brain activity imaging research using an fMRI (functional Magnetic Resonance Imaging) system (not only equipment but also research support). The fMRI system used is the Shimadzu-Picker MAGNEX ECLIPSE Power Drive 250 with 1.5 T static magnetic field. In June 2000, the service of the center will also be opened to the public for brain researchers. (Additional information: http://www.isd.atr.co.jp/baic/)
3 Research funded by governmental organizations
(7) Computational Neuroscience Project (PL: M.Kawato)
Computational neuroscience aims to understand brain functions deeply and essentially to the extent that the same functions be realized by a computer program or artificial machinery. We believe that the following three levels are essential for a complete understanding of brain functions: (a) hardware level; (b) information representation and algorithms; and (c) computational theory. We are studying high-level functions of the brain utilizing multiple methods such as neurophysiological modeling of the basal ganglia and cerebellum; psychophysical and behavioral analysis off visual motor learning; measurement of brain activity by fMRI and MEG study; mathematical analysis; and computer simulation of neural networks.
(8) 3-D Image Congnition Project (PL: S.Akamatsu)
The goal of this project is to make mathematical models for face recognition problems. Computer vision techniques are combined with psychological experiments to learn the relationships among faces, expressions, and emotions.
(9) Adaptive Visual Integration Project (JST PRESTO project)(PL: H.Anddo)
This project investigates computational mechanisms that adaptively recognize three-dimensional visual scenes, and estimate the future states of dynamically changing worlds based on the active processes of visual learning, synthesis, and prediction. Neural models of dynamic visual cognition and integration are developed and tested using experimental techniques of visual psychophysics and functional brain imaging.
(10) Neuroinformatics Project (PL: K.Doya)
The aim of the Neuroinformatics Project Iis to understand the neural mechanisms of learning and control of human behaviors based on recent advances in neurobiological studies and computational learning theory. A major research topic, supported by CREST, JST, is the development of 'metalearning' algorithms for adaptive agents and modeling of the neuromodulators, such as dopamine and serotonin, as mediums for metalearning in the brain.
(PL: Project Leader)