Lapas attēli
PDF
ePub
[merged small][merged small][ocr errors][merged small][merged small]

solutions exhaustively. The computation requires n! search paths, where n is the number of tracks. Even with pruning methods, the multitarget tracking problem is extremely computation intensive. Using a Hopfield network operating in the optimization mode to solve the traveling salesman problem, he is able to show that the Hopfield net can be used to solve the multitarget tracking problem in a short time. Furthermore, he suggests that with the rapid pace of development in neural network hardware VLSI technology, realtime multitarget tracking is soon to become a reality.

[blocks in formation]

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

TAIWAN

were used to insure terminal attractions in the backpropagation learning algorithm. Lipschitz conditions were used to guarantee convergence and stability. An algorithm has also been stability. An algorithm has also been developed to determine the appropriate number of hidden units in a multilayer perceptron network.

In Taiwan, much of the neural network work is centered at National Taiwan University and National Chiao Tung University. There is also a great deal of academic-industrial collaboration, such as between National Chiao Tung University and the Telecommunications Laboratory and the Indus- THAILAND trial Technology Research Institute.

The only work reported here from Thailand was that of Andreas Weigend at Chulalongkorn University on the dimension of the space of hidden units. The work was in collaboration with David Rumelhart while Weigend was at Stanford University. They showed that the effective number of parameters was changed during backpropagation training. The eigenvalue spectra of the covariance matrix of hidden units were analyzed to determine the ranks of the matrix. The technique was applied to the problem of prediction in the sun-spot time series. This work appeared to be quite advanced for a country like Thailand.

A very theoretical piece of work on fuzzy activation functions was presented by Chi-Cheng Jou of National Chiao Tung University. The fuzzy activation function generalized the two-value activation function in conventional perceptrons. The work was a mathematical analysis of the fuzzy neural networks started by Prof. B. Kosko at USC. Another theoretical work was that of Fu-Chuang Chen on the convergence properties of a modified backpropagation learning algorithm. The modification was in introducing a deadzone around the origin. He showed that the norm of the parameter error will converge to a constant provided the initial errors were sufficiently small. Also at National Chiao Tung Univer- SUMMARY sity, neural network algorithms were used on a multistage network to solve the traffic control problem.

At National Taiwan University, Hopfield nets with time-varying energy functions were used to solve the traveling salesman problem. The time-varying function was similar to the concept of system entropy. They did not, however, prove and guarantee the stability and convergence of the solution. Also at National Taiwan University, timevarying gains in the weight update laws

As can be seen from this report, there is a great deal of research and development work in the area of neural networks in Asia. Much of the work is in applying neural network technology to various problems. The applications are very diverse, but mostly in the area of pattern recognition. The most common neural network is the multilayer perceptron network with backpropagation learning algorithm. There is also a great deal of work in merging neural

network technology with fuzzy logic. Many of the researchers are educated in the United States or Great Britain. Clearly Japan is the most technologically advanced country in Asia. The number of researchers in the area of neural networks is large. With the start of the New Information Processing Technology program this year, the amount of effort is certain to increase dramatically. However, my impression is that much of the work is in applying neural network technology and that these Asian countries will continue to depend on the United States for theoretical results and basic research in neural networks.

Clifford Lau received his Ph.D. in electrical engineering and computer science from the University of California at Santa Barbara in 1978. He is presently a Scientific Officer in the Electronics Division at ONR and is responsible for the management of basic science research programs in VLSI algorithms and architecture for signal processing, VLSI reliability, ultra-dependable multiprocessor computers, and electronic neural networks. He is the editor of three books: An Introduction to Neural and Electronic Networks (with Zornetzer and Davis) by Academic Press, Neural Networks: Theoretical Foundations and Analysis by IEEE Press, and Artificial Neural Networks: Paradigms, Applications, and Hardware (with Sanchez-Sinencio) by IEEE Press.

SCIENCE STRUCTURE OF JAPANESE GOVERNMENT AND EXCHANGE POSSIBILITIES

This article provides an overview of organizations within the
Japanese Government that support science and describes the
sources of support for international exchanges.

ORGANIZATIONAL STRUCTURE OF SCIENCE AND TECHNOLOGY (S&T) IN JAPAN

The Government of Japan (GOJ) is organized around the Prime Minister, who heads the government. He is assisted by a Cabinet, composed of Ministers (Education, Justice, Foreign Affairs, etc.). Each of the ministries has its own budget. However, two ministries are especially important, the Ministry of International Trade and Industry (MITI) and the Ministry of Education (Monbusho). Both of these support significant scientific activities by a variety of mechanisms including grants for research and development (R&D), tax incentives, loans, and others. These organizations have subunits that also provide support. Monbusho supports the national universities and gives research grants, the latter somewhat like the National Science Foundation. Examples include the National Institute for Education Research, the National Education Center, and the National Women's Education Center. Monbusho also administers the National Center for Science Information System (NACSIS). MITI runs the Patent Office, which is highly automated. Some projects are supported by joint programs.

by Iqbal Ahmad and David K. Kahaner

To a lesser degree the Ministry of Post and Telecommunications (MPT) is also important.

In addition to the ministries, directly under the Prime Minister's office is the Science and Technology Agency (STA), which operates much like a ministry. STA, in addition to having research programs, also supports the Japan Information Center of Science and Technology (JICST).

The GOJ also includes its legislative arm, the Diet, which supports the National Diet Library (NDL), similar to the Library of Congress, in which all unclassified publications (including grey literature) published in Japan are deposited.

Within the Government, the highest policy making body for S&T is the Office of the Prime Minister, who is advised by the Science and Technology Council and the Science Council. These two councils establish national goals and provide broad directions for S&T and in general decide S&T issues of national importance. The membership of these councils consists of eminent scientists and outspoken science policy experts. These councils have a strong say in Japan's Federal S&T budget and have been responsible for the increase from ¥10,627B ($80B) in 1988 to more than ¥13T ($100B) in 1990.

MITI and Monbusho, along with STA, share the responsibility of planning, funding, and overseeing the government-sponsored S&T programs. The Federal Trade Commission governs the legality of the business practices and rules on the MITI plans and activities. The Ministry of Finance approves all the budget and justification for new programs and requirements of additional funds.

Science and Technology
Agency (STA)

The agency is located in the Prime Minister's Office and receives about 25% of the government S&T funds for major programs such as space and nuclear reactor projects. The agency also has the responsibility for promoting basic research in industry through the Japan R&D Development Corporation (JRDC) and supporting new initiatives such as the Exploratory Research on Advanced Technology (ERATO) program. Attached to STA are six research institutes:

• National Institute for Research in Inorganic Materials (NIRIM)

• National Research Institute for Metals (NRIM)

• National Aerospace Research Laboratories

• National Institute for Radiological Sciences

• National Research Center for the Disaster Prevention

• National Institute of Resources

Another function of this agency is to administer programs of research fellowships and a number of large national and international research programs. It is also responsible for managing research fellowships and grants. Of these the most well known is the Japan Society for the Promotion of Sciences (JSPS) fellowships. This is discussed in more detail in a later section.

Ministry of International Trade and Industry (MITI)

MITI receives about 13% of government S&T funds and has the responsibility for formulating industrial technology plans, providing subsidies and/ or funding, and selecting/persuading/ organizing participating industrial R&D groups/associations to work in collaboration with one or more of the 16 MITI national laboratories. These national laboratories are administered by the Agency of Industrial Science and Technology (AIST), which in 1985 had a budget of approximately $1B. A sister agency, called the Japan Industrial Technology Association (JITA), functions as a licensing agency of AIST and provides regular information on foreign technology developments. MITI relies on cooperative mechanisms with industry to leverage much more R&D than could be expected of the 13% Federal S&T funds allocated to it. The 16 national research laboratories managed by AIST are as follows:

• National Research Laboratory for Metrology

• Mechanical Engineering Laboratory for foreign scientists, etc. The execution and management of these research

National Chemical Laboratory for programs are shared by STA, MITI, Industry

• Government Industrial Research Institute, Osaka

• Government Industrial Research Institute, Nagoya

and an agency called the New Energy and Industrial Technology Development Organization (NEDO).

NEDO is a quasi-government body that was established in 1980 to promote coordinated development and commercialization of alternative energy sources for oil to reduce the depen

• Government Industrial Research dency of the Japanese economy on Institute, Kyushu

• Government Industrial Research Institute, Tohoku

• Government Industrial Research Institute, Shikoku

imported oil. But as of 1988, its functions have been expanded to include MITI's R&D projects at the national laboratories, large scale projects, development of advanced research facilities, and initiation of international joint research grant projects.

[blocks in formation]
[blocks in formation]

In addition to MITI, other ministries also have R&D activities, but at a smaller scale. They also have programs of science and technology exchange involving scientists from the developed countries.

The concept of "technopolis" materialized in Japan in the form of Tsukuba Science City. Recently, the Kansai district has decided to develop a science city in the Kansai area in the middle of the triangle made by the cities of Kyoto, Nara, and Osaka. In all, 26 such projects are planned throughout Japan. Also, Japan Defense Agency (JDA) there is a movement towards establishing research parks. One of these is being developed in the Kansai district. At this time both the technopolis and the research park concepts are getting a lukewarm response from industry. The main reasons are the large investment required to establish these research centers and the fact that the corporations are favoring centralizing their research activities in their own corporate laboratories (“Japan as Scientific and Technological Superpower 1990," by Justin Bloom, Department of Commerce Report PB90234923).

Industrial Research
Laboratories

As is well known, most of the high tech research is conducted by industry in Japan. Unlike the United States, where the industrial contribution to R&D investment is about 50% or less, in Japan industry's share is more than 80%. The attitude of industry towards accepting foreign researchers is also changing. The Tokyo office of the National Science Foundation has prepared a survey report in which the names of the companies willing to accept foreign research workers and the specified fields are summarized (Directory of Japanese Company Laboratories Willing to Receive American Researchers, 1 March 1991). According to a recent survey, in 1991, there were approximately 750 foreign research personnel employed by private corporations in Japan; 189 of them are from the United States.

National Science Foundation

In the context of U.S.-Japan S&E exchange, the most important body involved is the National Science Foundation (Japan office). This office not only manages the program under which U.S. scientists and students visit Japan, both for short and long periods, but also supports JSPS programs of Monbusho and some international programs operated by both STA and AIST.

INTERNATIONAL
COOPERATION IN R&D AND
OPPORTUNITIES FOR
U.S. SCIENTISTS

There are four major research insti-
tutes engaged in defense-related proj-
ects in Japan. These institutes employ
a total of 1,000 persons. Compared
with U.S. Department of Defense
(DOD) R&D laboratories, the JDA
activity is very small. At the same time
there is no formal Memorandum of
Understanding (MOU) between the
United States and Japan to establish a
scientist/engineer (S&E) exchange.
Examples of some U.S. DOD scientists
working for a short period at the Tech-
nical Research and Development Insti-
tute (TRDI) of JDA or a Japanese
scientist working in the United States
are few and far between. Recently there
has been a considerable updating of
the research facilities at these institu-
tions and some mutually agreeable high
tech research projects could be under-
taken at these laboratories. However,
that will be possible only if a formal
MOU is established between DOD and AIST/NEDO
JDA.

R&D Activities of
Private Corporations

Over 14% of Japanese corporations have R&D facilities outside Japan. A majority of these R&D facilities have been established for product development matched to local needs. Reportedly 71 companies from the United States have R&D facilities in Japan, mainly for developing products specifically to meet the needs of Japan.

Discussions with responsible officials in MITI and NSF indicated that there is no reliable source of information on the overall international cooperative R&D activities including data about the number of foreign scientists in Japan versus the number of Japanese scientists abroad. A MITI official stated that MITI was conducting a survey on this issue. The report will be ready by the end of March 1992. The data summarized in the following were obtained from discussions and brochures available on some of the programs.

The following programs are managed by NEDO.

Foreign Researcher Invitation Program. Under the program established in 1988, AIST invites foreign scientists for 6-12 months to conduct research at the 16 AIST laboratories. In 1989, AIST assigned NEDO to administer this program. In 1990, of the 30 scientists invited, 3 were from the United States. In 1992, this number may increase to 50. NEDO provides the expenses for air transportation to and from Japan,

« iepriekšējāTurpināt »