IEEE/Nagoya University World Wisemen/women Workshop(WWW) On Fuzzy Logic and Neural Networks/Genetic Algorithms

[Dr. David Kahaner is a numerical analyst on sabbatical to the Office of Naval Research-Asia (ONR Asia) in Tokyo from NIST. The following is the professional opinion of David Kahaner and in no way has the blessing of the US Government or any agency of it. All information is dated and of limited life time. This disclaimer should be noted on ANY attribution.] [Copies of previous reports written by Kahaner can be obtained using anonymous FTP from host cs.arizona.edu, directory japan/kahaner.reports.]

From:
Dr. David K. Kahaner
US Office of Naval Research Asia
(From outside US): 23-17, 7-chome, Roppongi, Minato-ku, Tokyo 106 Japan
(From within US): Unit 45002, APO AP 96337-0007
Tel: +81 3 3401-8924, Fax: +81 3 3403-9670
Email: kahaner@cs.titech.ac.jp
Re: Workshop on Fuzzy, Neural, Genetic Alg, Nagoya, 8/94
09/12/94 (MM/DD/YY)
This file is named "www.94"

The following was prepared by Jim Toerresen <Adresses removed >

1994 IEEE/Nagoya University World Wisemen/women Workshop(WWW) On Fuzzy Logic and Neural Networks/Genetic Algorithms, 9-10 Aug 94, Nagoya, Japan

This was the third workshop in a series organized to cover the field of research, combining Fuzzy & Neuro and Fuzzy & Genetic Algorithms (GA). A difficulty associated with fuzzy logic is knowledge acquisition from experts and/or finding knowledge from unknown tasks. Neural networks and genetic algorithms are gaining attention for their potential in the area of efficient knowledge acquisition/finding. This was the background for organizing the workshop. It was held in the nice Nagoya Symposion at Nagoya University and was well organized by the general chair Prof. Takeshi Furuhashi.

Dr. Takeshi Furuhashi
Associate professor
Department of Information Electronics
Nagoya University
Foro-cho, Chikusa-ku, Nagoya 464-01 JAPAN
Tel: +81 5278 9 2792; Fax: +81 5278 9 3166
Email: FURU@BIOELE.NUEE.NAGOYA-U.AC.JP

Of more than 50 submitted papers, 19 papers were accepted for presentation. Of those, 11 were from abroad. The first day was devoted primarily to presentations related to fuzzy-neuro, while fuzzy-GA was covered the second day. Most workshop attendees, had also been to the IIZUKA'94 conference [a much larger conference on related topics that was held earlier this summer -- I was not able to attend, DKK]. The workshop was opened by the president of Nagoya University, Prof. Nobuo Kata followed by a short talk from Prof. Yoshiki Uchikawa in the advisory committee. He also belongs the Dept. of Information Electronics at Nagoya University. He used the word soft computing as a global term for this field of research. The workshop was a mix of theory and applications. The papers were not that much connected to each other thus below I will give a description of some of the papers presented. I will mainly refer to the papers presented by researchers from Asia.

"Conceptual-level-learning-based Refinement of a Human-motion Model" Presented by: Hirohide Ushida,
Lab. for International Fuzzy Engineering Research, Japan
Email: USHIDA@FUZZY.OR.JP

Human motion models are important in virtual reality applications. The problem is how to model this information. The presentation introduced conceptual level learning (CLL) which imitates the way humans are learning i.e. through macro instructions. The system receives input either as motion or words from a training person. This information is transformed and stored as fuzzy-based knowledge (fuzzy associative memory). A tennis motion system was used in their experiments. A camera detected human motion using indicators on the body of person playing. Several different tennis-strokes were stored in the system. Then strokes from other tennis players were input to test the recognition robustness. They obtained 84.2 % correct recognition (average for forehand, backhand and smash strokes).

"A Fuzzy Gaussian Potential Neural Network Based on Input-data-Functional Reasoning"
Presented by: Mohammad Teshnelab,
Saga University, Japan

The paper presents the design of a fuzzy gaussian potential neural network to be used e.g. as a fuzzy controller for robotic manipulators. Each fuzzy membership function consists of a gaussian potential functions (GPF). The method presented benefit from that if the number of inputs to the system increases, the number of fuzzy rules required does not increase. Thus the total training time is shortened compared to other method.

"Information Aggregation Networks based on Fuzzy Integral"
Presented by: Keon-Myung Lee
Korea Advanced Institute of Science and Technology, Korea
Email: KMLEE@MONAMI.KAIST.AC.KR

In pattern recognition (and other decision making problems) it is often necessary to obtain an overall evaluation from several partial evaluations. The paper describes an extension of Sugeno's Fuzzy Integral, which is a method to aggregate partial evaluation values with respect to the degree of importance of the evaluation items. In the proposed method, genetic algorithms is used to identify fuzzy measures values for the importance degrees. Several other improvements are also introduced. The preference degree of secondhand cars was applied to show the usefulness of the method.

Keynote Lecture: "Chaos in Neuro-Fuzzy Systems and Modeling Issues"
Presented by: H.N.L. Teodorescu

This speech was similar to the one given at IJCNN'93 [International Joint Conference on Neural Networks, held in Nagoya Oct 1993 and summarized in my report "ijcnn.93" 28 Dec 1993, DKK]. He started off with an introduction to neuro-fuzzy systems. The modeling issue was a main part of the talk as he outlined that rules driven decision best was modeled by fuzzy logic systems, while experience driven decision needs neural network systems. Then, by combining these two systems, we should be able to get a complex model that better mimic the human-related processes. Then he went on with describing different discrete-time recurrent neuro-fuzzy systems.

"A New Approach to Genetic Based Machine Learning and an Efficient finding of Fuzzy Rules"
Presented by: T. Furuhashi (Workshop chairman)
Email: FURU@BIOELE.NUEE.NAGOYA-U.AC.JP

The busy chairman also had time to give a presentation of the "Nagoya approach". The algorithm is a new approach for Genetic Based Machine Learning (GBML) and is applied to find complex fuzzy rules for obstacle avoidance of mobile robots. The idea is based on a hypthesis of evolution, i.e., most cells having mutations will die. However, if a cell survives and produce a large number of similar cells in the organ, the new genes may propagate to the sexual organs and will be inherited to the offspring.

"A paper Currency Recognition Method by a Neural Network Using Masks and Mask Optimization by GA"
Presented by: Fumiaki Takeda
Glory Ltd, Japan

The presentation was on adapting GA to find effective masks for paper currency recognition (by using masking, characteristic parts of an image can be emphasized). The recognition itself is done by a perceptron neural network. GA is used to find the optimal position of the masks to obtain both shortened training time and to improve the generalization. The company (Glory) has developed a neuro-board which adopts this technique. They claim it is 10 times faster than conventional recognition machines. [Glory also described this work at IJCNN'93 (see remark above), DKK]

The seminar was ended by a panel discussion which was introduced by Hideyuki Takagi from Matsushita Electric Industrial Co, Ltd. He gave a similar presentation as the one given at IJCNN'93 in i Nagoya last year. One comment from the panel was that new theory ought to be tested on realistic problems.

Other papers at the workshop: