Lapas attēli


solutions exhaustively. The computation

requires n! search paths, where n is the Only research by Professor Due to the fact that the conference number of tracks. Even with pruning G.G. Coghill and his students at the was held in Singapore, a fair number of methods, the multitarget tracking probUniversity of Auckland was presented papers were presented. Research is lem is extremely computation intenat this conference. There a multilayer concentrated mostly at the National sive. Using a Hopfield network operatperceptron network is trained using University of Singapore, as shown in ing in the optimization mode to solve error backpropagation to recognize the Table 6.

the traveling salesman problem, he is angle and length representations of Of particular interest to the Navy is able to show that the Hopfield net can various shapes in robotic systems. A the work of Leonard Chin at Nanyang be used to solve the multitarget trackcompetitive learning network is used Technical University on a neural net- ing problem in a short time. Furtherto estimate the centroid of clusters of work approach in multiple target track- more, he suggests that with the rapid patterns. Also, a method has been devel- ing. Conventional multitarget tracking pace of development in neural netoped for selecting the optimum oper- uses either Kalman filtering or the work hardware VLSI technology, realating point for a network consisting of Dempster/Shafer method. These meth- time multitarget tracking is soon to hardlimiting neurons, for use in con- ods try to find the maximum probabil- become a reality. tent addressable memories.

ity solution to search all possible

[blocks in formation]

Parallel implementation on transputers
Processing of medical images
Character recognition
Forward generating neural network
Competitive unsupervised learning
Neural network expert system
Nearest neighbor learning
Neural net control system
Robot motion control
Electrical load forecasting
Multiple target tracking
Cause associator network
Forecasting of electricity consumption
Backpropagation for prediction
Controller in the presence of disturbances
Fuzzy feature extraction
Fuzzy neural system for decision making
Dual network expert system
Case-based diagnostic expert system
Forward kinematic problem
Robust stability of analog VLSI
Training algorithm for fast error propagation
Inverse nonlinear control
Rotation invariant neocognitron
Unlearning in BAM
Occluded object recognition

National Univ. of Singapore
Acoustical Services Pte Ltd.
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
Nanyang Tech. Univ.
Human Interface Engin. Pte Ltd.
National Univ. of Singapore
Information Tech. Inst.
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
National Computer Board
National Univ. of Singapore
National Univ. of Singapore
Nanyang Tech. Univ.
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore
National Univ. of Singapore


were used to insure terminal attrac- network technology with fuzzy logic.

tions in the backpropagation learning Many of the researchers are educated In Taiwan, much of the neural net algorithm. Lipschitz conditions were in the United States or Great Britain. work work is centered at National used to guarantee convergence and

used to guarantee convergence and Clearly Japan is the most technologi- . Taiwan University and National Chiao stability. An algorithm has also been cally advanced country in Asia. The Tung University. There is also a great developed to determine the appropri- number of researchers in the area of deal of academic-industrial collabora- ate number of hidden units in a multi- neural networks is large. With the start tion, such as between National Chiao layer perceptron network.

of the New Information Processing Tung University and the Telecommu

Technology program this year, the nications Laboratory and the Indus- THAILAND

amount of effort is certain to increase trial Technology Research Institute.

dramatically. However, my impression A very theoretical piece of work on The only work reported here from is that much of the work is in applying fuzzy activation functions was presented Thailand was that of Andreas Weigend neural network technology and that by Chi-Cheng Jou of National Chiao at Chulalongkorn University on the these Asian countries will continue to Tung University. The fuzzy activation dimension of the space of hidden units.

dimension of the space of hidden units. depend on the United States for theofunction generalized the two-value The work was in collaboration with retical results and basic research in neural activation function in conventional David Rumelhart while Weigend was networks. perceptrons. The work was a mathe- at Stanford University. They showed matical analysis of the fuzzy neural that the effective number of param

Clifford Lau received his Ph.D. in networks started by Prof. B. Kosko at eters was changed during backpropa- electrical engineering and computer USC. Another theoretical work was gation training. The eigenvalue spec- science from the University of that of Fu-Chuang Chen on the con- tra of the covariance matrix of hidden California at Santa Barbara in 1978. vergence properties of a modified back- units were analyzed to determine the He is presently a Scientific Officer in propagation learning algorithm. The ranks of the matrix. The technique was the Electronics Division at ONR and modification was in introducing a dead- applied to the problem of prediction in is responsible for the management zone around the origin. He showed the sun-spot time series. This work of basicscience research programs in that the norm of the parameter error appeared to be quite advanced for a VLSI algorithms and architecture for will converge to a constant provided country like Thailand.

signal processing, VLSI reliability, the initial errors were sufficiently small.

ultra-dependable multiprocessor Also at National Chiao Tung Univer- SUMMARY

computers, and electronic neural sity, neural network algorithms were

networks. He is the editor of three used on a multistage network to solve As can be seen from this report, books: An Introduction to Neural and the traffic control problem.

there is a great deal of research and Electronic Networks (with Zornetzer At National Taiwan University, development work in the area of neural and Davis) by Academic Press, Neural Hopfield nets with time-varying energy networks in Asia. Much of the work is Networks: Theoretical Foundations and functions were used to solve the travel- in applying neural network technology Analysis by IEEE Press, and Artificial ing salesman problem. The time-varying to various problems. The applications Neural Networks: Paradigms, , function was similar to the concept of are very diverse, but mostly in the area Applications, and Hardware (with system entropy. They did not, however, of pattern recognition. The most com- Sanchez-Sinencio) by IEEE Press. prove and guarantee the stability and mon neural network is the multilayer convergence of the solution. Also at

perceptron network with backpropaNational Taiwan University, time- gation learning algorithm. There is also varying gains in the weight update laws a great deal of work in merging neural



This article provides an overview of organizations within the
Japanese Government that support science and describes the

sources of support for international exchanges.

by Iqbal Ahmad and David K. Kahaner


To a lesser degree the Ministry of Post MITI and Monbusho, along with STRUCTURE OF SCIENCE and Telecommunications (MPT) is also STA, share the responsibility of planAND TECHNOLOGY (S&T) IN important.


ning, funding, and overseeing the JAPAN

In addition to the ministries, directly government-sponsored S&T programs.

under the Prime Minister's office is the The Federal Trade Commission govThe Government of Japan (GOJ) is Science and Technology Agency (STA),

Science and Technology Agency (STA), erns the legality of the business pracorganized around the Prime Minister, which operates much like a ministry. tices and rules on the MITI plans and who heads the government. He is assisted STA, in addition to having research activities. The Ministry of Finance by a Cabinet, composed of Ministers

programs, also supports the Japan Infor

the Japan Infor- approves all the budget and justifica(Education, Justice, Foreign Affairs, mation Center of Science and Tech- tion for new programs and requireetc.). Each of the ministries has its own nology (JICST).

ments of additional funds. budget. However, two ministries are The GOJ also includes its legislaespecially important, the Ministry of tive arm, the Diet, which supports the Science and Technology International Trade and Industry (MITI) National Diet Library (NDL), similar Agency (STA) and the Ministry of Education to the Library of Congress, in which all (Monbusho). Both of these support unclassified publications (including grey The agency is located in the Prime significant scientificactivities by a vari- literature) published in Japan are Minister's Office and receives about ety of mechanisms including grants for deposited.


25% of the government S&T funds for research and development (R&D), tax Within the Government, the highest major programs such as space and incentives, loans, and others. These policy making body for S&T is the Office nuclear reactor projects. The agency organizations have subunits that also of the Prime Minister, who is advised also has the responsibility for promotprovide support. Monbusho supports by the Science and Technology Coun- ing basic research in industry through the national universities and gives cil and the Science Council. These two the Japan R&D Development Corporesearch grants, the latter somewhat councils establish national goals and ration (IRDC) and supporting new like the National Science Foundation. provide broad directions for S&T and initiatives such as the Exploratory Examples include the National Insti- in general decide S&T issues of national Research on Advanced Technology tute for Education Research, the importance. The membership of these (ERATO) program. Attached to STA National Education Center, and the councils consists of eminent scientists are six research institutes: National Women's Education Center. and outspoken science policy experts. Monbusho also administers the National These councils have a strong say in • National Institute for Research in Center for Science Information Sys- Japan's Federal S&T budget and have Inorganic Materials (NIRIM) tem (NACSIS). MITI runs the Patent been responsible for the increase from Office, which is highly automated. Some ¥10,627B ($80B) in 1988 to more than • National Research Institute for projects are supported by joint programs. ¥13T ($100B) in 1990.

Metals (NRIM)

• National Aerospace Research • Mechanical Engineering Laboratory for foreign scientists, etc. The execuLaboratories

tion and management of these research • National Chemical Laboratory for programs are shared by STA, MITI, • National Institute for Radiological Industry

and an agency called the New Energy Sciences

and Industrial Technology Develop

Government Industrial Research ment Organization (NEDO). • National Research Center for the Institute, Osaka

NEDO is a quasi-government body Disaster Prevention

that was established in 1980 to pro• Government Industrial Research mote coordinated development and • National Institute of Resources Institute, Nagoya

commercialization of alternative energy

sources for oil to reduce the depenAnother function of this agency is • Government Industrial Research dency of the Japanese economy on to administer programs of research Institute, Kyushu

imported oil. But as of 1988, its funcfellowships and a number of large

tions have been expanded to include national and international research • Government Industrial Research MITI's R&D projects at the national programs. It is also responsible for Institute, Tohoku

laboratories, large scale projects, develmanaging research fellowships and

opment of advanced research facilities, grants. Of these the most well known is • Government Industrial Research and initiation of international joint the Japan Society for the Promotion of Institute, Shikoku

research grant projects. Sciences (JSPS) fellowships. This is discussed in more detail in a later section. • Government Industrial Research Ministry of Education,

Institute, Chukoku

Science, and Culture Ministry of International Trade

(Monbusho) and Industry (MITI) • Fermentation Research Institute

This ministry manages 47% of the MITI receives about 13% of gov- • Research Institute for Polymers and government research funds, which are ernment S&T funds and has the respon- Textiles

distributed amongst the universities and sibility for formulating industrial tech

national centers for scientific research. nology plans, providing subsidies and/ • Geological Survey of Japan University research is exclusively supor funding, and selecting/persuading/

ported out of these funds. There is very organizing participating industrial R&D • Electrotechnical Laboratory little university-industry interaction. groups/associations to work in collab

Therefore, industry does not support oration with one or more of the 16 • Industrial Products Research much research in the universities. This MITI national laboratories. These Institute

is one of the often quoted reasons for national laboratories are administered

the poor state of university facilities by the Agency of Industrial Science and • National Research Institute of Pol- and research laboratories. Technology (AIST), which in 1985 had lution and Resources abudget of approximately $1B. Asister

Key Technology Center (KTC) agency, called the Japan Industrial • Government Industrial DevelopTechnology Association (JITA), func- ment Laboratory, Hokkaido

Its primary objective is to promote tions as a licensing agency of AIST and

research and development through provides regular information on foreign AIST is one of the most important industrial consortia composed of more technology developments. MITI relies elements of MITI. It implements its than one corporation by providing on cooperative mechanisms with indus- basic policy of promoting techno- investment funds. KTC also makes loans try to leverage much more R&D than globalism to stimulate scientific and to individual corporations to conduct could be expected of the 13% Federal technological creativity as well as dis- research on key basic technologies S&T funds allocated to it. The 16 tribution and transfer of the results of needed by various regions of the counnational research laboratories managed these activities. AIST is responsible for try. It is a financing body and does not by AIST are as follows:

providing funds for the international have R&D facilities of its own. The

research programs such as the Human center, with a budget of $186M, is jointly • National Research Laboratory for Frontier Science Program, international supported by MITI, industry, and the Metrology

fellowships, subsidies and invitations local prefectures.

Technopolis/Research Parks

Other Ministries

National Science Foundation

The concept of "technopolis” mate- In addition to MITI, other minis- In the context of U.S.-Japan S&E rialized in Japan in the form of Tsukuba tries also have R&D activities, but at a exchange, the most important body Science City. Recently, the Kansai dis- smaller scale. They also have programs

smaller scale. They also have programs involved is the National Science Fountrict has decided to develop a science of science and technology exchange dation (Japan office). This office not city in the Kansai area in the middle of involving scientists from the developed only manages the program under which the triangle made by the cities of Kyoto, countries.

U.S. scientists and students visit Japan, Nara, and Osaka. In all, 26 such proj

both for short and long periods, but ects are planned throughout Japan. Also, Japan Defense Agency (JDA) also supports JSPS programs of there is a movement towards establish

Monbusho and some international ing research parks. One of these is being There are four major research insti- programs operated by both STA and developed in the Kansai district. At tutes engaged in defense-related proj- AIST. this time both the technopolis and the ects in Japan. These institutes employ research park concepts are getting a a total of 1,000 persons. Compared INTERNATIONAL lukewarm response from industry. The with U.S. Department of Defense

with U.S. Department of Defense COOPERATION IN R&D AND main reasons are the large investment (DOD) R&D laboratories, the JDA OPPORTUNITIES FOR required to establish these research activity is very small. At the same time U.S. SCIENTISTS centers and the fact that the corpora- there is no formal Memorandum of tions are favoring centralizing their Understanding (MOU) between the Discussions with responsible offiresearch activities in their own corpo- United States and Japan to establish a cials in MITI and NSF indicated that rate laboratories (“Japan as Scientific scientist/engineer (S&E) exchange. there is no reliable source of informaand Technological Superpower 1990,” Examples of some U.S. DOD scientists tion on the overall international coopby Justin Bloom, Department of working for a short period at the Tech- erative R&D activities including data Commerce Report PB90234923). nical Research and Development Insti- about the number of foreign scientists

tute (TRDI) of JDA or a Japanese in Japan versus the number of Japanese Industrial Research

scientist working in the United States scientists abroad. A MITI official stated Laboratories

are few and far between. Recently there that MITI was conducting a survey on

has been a considerable updating of this issue. The report will be ready by As is well known, most of the high the research facilities at these institu- the end of March 1992. The data sumtech research is conducted by industry tions and some mutually agreeable high marized in the following were obtained in Japan. Unlike the United States, tech research projects could be under- from discussions and brochures availwhere the industrial contribution to taken at these laboratories. However, able on some of the programs. R&D investment is about 50% or less, that will be possible only if a formal in Japan industry's share is more than MOU is established between DOD and AIST/NEDO 80%. The attitude of industry towards JDA. accepting foreign researchers is also

The following programs are manchanging. The Tokyo office of the R&D Activities of

aged by NEDO. National Science Foundation has pre- Private Corporations pared a survey report in which the names

Foreign Researcher Invitation Proof the companies willing to accept foreign Over 14% of Japanese corporations gram. Under the program established research workers and the specified fields have R&D facilities outside Japan. A in 1988, AIST invites foreign scientists are summarized (Directory of Japanese majority of these R&D facilities have for 6-12 months to conduct research at Company Laboratories Willing to Receive been established for product develop- the 16 AIST laboratories. In 1989, AIST American Researchers, 1 March 1991). ment matched to local needs. Report- assigned NEDO to administer this According to a recent survey, in 1991, edly 71 companies from the United program. In 1990, of the 30 scientists there were approximately 750 foreign States have R&D facilities in Japan, invited, 3 were from the United States. research personnel employed by pri- mainly for developing products specif- In 1992, this number may increase to vate corporations in Japan; 189 of them ically to meet the needs of Japan. 50. NEDO provides the expenses for are from the United States.

air transportation to and from Japan,

« iepriekšējāTurpināt »