Conference heron::euro_swas_ai

379.0. "FWD: fyi: aimac.91" by MR4DEC::ROACH (TANSTAAFL !) Tue Oct 15 1991 00:48

                  I N T E R O F F I C E   M E M O R A N D U M

                                        Doc. No:  005816
                                        Date:     14-Oct-1991 03:58pm EDT
                                        From:     BEANE
                                                  BEANE@SHLACT@SELECT@MRGATE@NRGATE@NRO
                                        Dept:      
                                        Tel No:    

TO:  PATRICK ROACH@LMO


Subject: FWD: fyi: aimac.91                                          

From:	DECWRL::"[email protected]" "Jim Sims" 26-SEP-1991 20:57:03.35
To:	distribution:;@enet-gw.pa.dec.com (see end of body)
CC:	
Subj:	fyi

  [Dr. David Kahaner is a numerical analyst visiting Japan for two-years
   under the auspices of the Office of Naval Research-Asia (ONR/Asia).  
   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 from
   host cs.arizona.edu using anonymous FTP.]

To: Distribution
From: David K. Kahaner, ONR Asia [[email protected]]
Re: Artificial Intelligence based Measurement and Control, Kyoto Sept 91.
23 Sept 1991
This file is named "aimac.91"

ABSTRACT. A summary of the 8th International Symposium on Artificial 
Intelligence Based Measurement and Control (AIMaC'91), 12-16 Sept 1991, 
Kyoto Japan is given.

INTRODUCTION. AIMaC'91 is a spin-off meeting. Earlier this month (5-10 
September), the International Measurement Confederation (IMEKO) held a 
large World Congress XII in Beijing. A special interest group, the
Technical Committee on Measurement Theory (TC7) sponsored AIMaC'91.
Locally, the meeting was organized (and held at Ritsumeikan University
in Kyoto) by the Society of Instrument and Control Engineers of Japan
(SICE), a body with 9000 plus members. This was the 8th AIMaC symposium;
the earlier meetings give a fair sense of the center of gravity of the
membership.

        1969    Czechoslovakia
        1971    Hungary
        1973    GDR
        1975    The Netherlands
        1976    United Kingdom
        1978    USSR
        1979    USSR
        1981    Yugoslavia
        1982    GDR
        1984    Italy
        1986    FRG
        1987    Hungary
        1990    FRG

Thus, until this year, the activities of TC7 were essentially all 
European. The TC7 Committee that plans these meetings is composed of
about 20 scientists from 18 countries including the US.  This year's
meeting saw about 150 scientists from 21 countries come together in
Kyoto Japan, to listen to 6 plenary lectures, 10 invited papers and more
than 75 submitted papers. (About half the participants were local,
Japanese.) There was also a very small vendor exhibit, with about a
dozen displays showing a variety of test equipment and a fuzzy hardware
system for the appraisal of orthodontic results from Rohm.  Other than
myself, I was only able to locate one other US participant.  The main
topic of the current symposium was intelligent measurement, i.e., using
the computational power of today's smart chips to develop "intelligent
sensors", and to utilize AI techniques in modelling, design, management,
operation, diagnosis, etc.  A list of papers along with their authors,
is being prepared and will be distributed shortly. I have one copy of
the proceedings (all in English) and will attempt to provide copies of
selected reports to requestors.  Further details about this symposium,
as well as forthcoming ones, can be obtained from
        Prof Komyo Kariya
        AIMaC'91 Executive Committee
        Faculty of Science and Engineering
        Ritsumeikan University
        56-1 Tojiin-kita, Kita-ku, Kyoto, Japan
         Tel: +81-75-465-1111, Fax: +81-75-465-1209


COMMENTS.
(1) The distinction between sensors and computers is narrowing, and in 
some situations might have already disappeared. Several perfect examples 
of this were given by 
	Shoei Kataoka 
	VP Research
	Sharp Corp, Corporate R & D Group
	273-1 Kashiwa, Kashiwa-shi, Chiba 277 JAPAN
who gave one of the Plenary lectures, and described two of his company's
research projects.  (Kataoka also commented that part of the research
was funded by a recently ended 10-year MITI project, Fundamental
Technology for the Next Generation, and that the type of research and
results proved that "it was untrue that Japanese research is only in
applied R&D areas.")

Kataoka focused on what he termed "sensible" sensors. These are devices
that combine physical sensors, actuators that incorporate feedback 
from/to the sensors, and a module that allows deduction, learning and 
abstraction. Such a sensor is linked to a large knowledge data-base and 
also to a human (perhaps remotely). Kataoka hopes that this type of 
sensor could be used to detect such things as emotion by examining a 
real time image of a person and extracting information about tension,
anger, joy, age, etc.  Of course we are not at that stage yet, but the
sensors Kataoka showed illustrated how far we have come from the days of
analog voltmeters.

(a) 3D Image Sensor on a chip. Kataoka made a very simple point that,
without dismissing the unsolved technical problems, two dimensional ICs
are bound to be less efficient than three dimensional ones by showing
photographs of two Sharp facilities, one a sprawling factory in
Tennessee, and another an office building in Tokyo; the latter has
shorter paths, higher packing density, is multi-function, etc. Sharp is
now working on their second prototype of a four layer chip fabricated by
SOI (silicon on insulator) technology 14.3mm^2, with about 280K devices.
The top layer (fourth) has 55,000 photo diodes and acts as an imager,
i.e., an optical to electronic conversion (what used to be the
"sensors") of 420 pixels per character. This represents a 10x14 bit
character matrix.  Signals are passed in parallel to the third layer
that digitizes, using majority decision, and then passes the data (also
in parallel) to the second layer which acts as a data mask. The bottom
(first) layer is an associative memory.  If an input image is thought to
match one of the stored images then the latter image is output. The
individual layers are zone melted polysilicon (into a single crystal),
although the circuit design on the chip is basically made according to
the design method used for conventional 2D ICs. Kataoka is hoping that a
real 3D architecture will be designed in future.  Nevertheless, the chip
works as we were shown several examples of input and output characters.
(In the context of this discussion it is not appropriate to put too much
emphasis on the algorithmic details.) Is this a sensor or a computer?

(b) Intelligent optical odor sensor. This is also an ongoing project, and 
is being supported as part of MITI's program in bio-electronics. Odor 
(smell sense) has been one of the most difficult sensors to realize.  
Partly this is because such a sensor must be exceptionally sensitive, and 
partly because smell is a combination of various kinds of molecules. The 
human olfactory system can detect very low levels of odors, and also 
distinguish different odors present in mixtures. The human system 
consists of several million receptor cells and olfactory bulbs. The 
receptor cells sense odors and their signals are transmitted to olfactory 
bulbs where signals are parallel processed and some profile of the signal 
is used to distinguish odors. Sharp's approach to such a sensor uses 
photo sensitive organic films. (These films change color or intensity on 
contact with different molecules.) The receptor plate consists of a two
dimensional array of various dye films (the dyes used are
solvatochromic dyes and color formers). Output of the plate goes to a
photochromic film which acts like an optical storage. Finally output
goes to a third layer, of thresholding film, acting like an optical
processor. The films were prepared on glass by spin coating from
tetrahydrofuran or other solutions containing dye and a polymer. In the
example Kataoka showed that various gases could be distinguished
(methanol, acetone, isoamyl acetate, acetonitorile, ethanol, ammonia,
and others). At the moment this is a pure optical device, i.e.,
non-digital. But Sharp is now experimenting with attaching the output to
a neural network to enhance its decision making ability. Is this a
sensor or a computer?

(2) Scope of the papers. As might be expected at a conference such as 
this one, the technical range of the talks was very wide. There were 
several very high level papers. One of the best examples of these was 
given by 
	Ludwik Finkelstein 
	Measurement and Instrumentation Center
	City University
   	London EC1V OHB, United Kingdom
in which information, knowledge, measurement, perception, information
machines, etc., were given precise definitions in the context of sensor
and measurement applications.

	Gerhard Barth 
	German Research Center for AI
	Postfach 2090, D-6750 Kaiserslautern, GERMANY
gave a very readable and coherent survey explaining what is knowledge
processing, what are neural networks, etc. (This was an excellent
introduction, but the content was a bit disappointing based on his title
"Knowledge Based Systems--State of the Art and Future Trends").  For me
though, the most interesting of these general talks, and the one that
defined best the direction this field is heading, was by
	Giuseppe Zingales 
        Electrical Engineering Department
	Via G.Gradenigo 6/A I-35131
	University of Padua, Padova, Italy
on "The Role of Artificial Intelligence in Measurement".  Zingales
points out, among other things, that "When the outcome of some procedure
depends on the processing within a logical structure of physical
quantities, attention should be paid to the real meaning of those
measurements, that is to their logical foundations. Any AI application
needs to lay on a sound theoretical basis; providing the required
support means that basic aspects in the science and practice of
measurement have to be thoroughly understood, or, in some cases, even
revised and reassessed." His paper also includes an bibliography of
several dozen items.
	Valery Ivanor 
	All-Union Research Inst of Electrical Measuring Instruments
	85, Prosveshenia Av, Leningrad, USSR 
and 
	Geniy Kavalerov 
	All-Union Scientific Technical Society of Instrument Industry
	17, Marx Av, Moscow, USSR
also gave a high level formalization of metrological aspects of
information processing.

Another remarkable survey, although not related to AI, was by 
	Jiaoxian Ning 
	Institute of Mechanics
	Chengdu University of Science and Technology
	Chengdu 610065, Sichuan, PRC
"A Decade of Development of Electric Measurement in Theory and Applied
Technology in China". Although this is far from my area of technical
expertise, it is written in exceptionally clear English and has a 41
item bibliography.  Another overview by
	Dietrich Hofmann
	Friedrich-Schiller University Jena
	Lobdergraben 32, D(O)-6900 Jena, GERMANY
in the field of PC-based visualization techniques for in-plant quality
control. Very nicely done, with a useful list of unsolved problems, but
not about AI. Another excellent tutorial on "scalelogy," "nominal
scale", and classification, such as what one ought to  use to analyzing
IQ scores, by
	Hiroshi Watanabe
	Faculty of Engineering
	Tokyo Institute of Technology
	2-12-1, Oh-Okayama, Meguro-ku, Tokyo 152 JAPAN.  
This looks to be fertile ground for AI approaches, although the author
only touched on these very briefly.

A large number of papers were applications, on topics that varied all
over the place, from the CAD/CAM area, boiler operation, driving
simulator, flow regulators, etc.  Most of these paid very little
attention to what computer scientists would call AI in the sense that a
methodology was developed or applied that could be generalized and used
later, although some authors did employ fairly well established methods
such as case based reasoning, model-based methods, fuzzy theory, system
identification via spectral analysis, recursive estimation, AR models,
maximum likelihood, image processing, etc. Several of these papers were
very interesting in their own right, such as one by
	Takatomo  Mori
	Dept of Computer Science and Systems Engineering
	Faculty of Science and Engineering
	Ritsumeikan Univeristy
	56-1, Tojiin-kita, Kita-ku, Kyoto 603 JAPAN
who has developed a fast extension to standard Hough transforms and
applied this to extracting road edges from the visual field of a mobile
robot.  Road-edge determination was also a key aspect of a paper by
	Yoshikuni Okawa
	Faculty of Engineering
	Osaka University, Yamadaoka, Suita, Osaka, JAPAN
in the (much more difficult) problem of reading signs painted on road
surfaces.  The one US paper by
	Chun Cho 
	Fisher Controls International
	205 S. Center Street
	Marshalltown, Iowa 50158, USA
was about optimization of energy in industrial plants.  This uses
interesting optimization techniques, but as far as I can tell no AI.

There were a few good applications of expert systems, by now pretty well 
understood in the engineering community. For example 
	Shin Tanabe
	AI Eng Section, Application Engineering Department
	Yokogawa Electric
	Shihjuku Center Buliding (50F)
        1-25-1, Nishi-Shinjuku-ku, Tokyo 163, JAPAN
described a real time system that his company uses to run one of their
power plants. Another paper with somewhat more generality was given by
	David Vander 
	Ford Motor Company of Australia
	Melbourne, AUSTRALIA
describing a system to assist designers in selecting measuring
instruments depending on the surface finish, flatness, dimensional
accuracy required, etc. Biesen (Belgium) gave a good summary of PC based
expert systems designed to help experimentalists. To my surprise I had
used two of these, Asyst, and Asystant and never realized that they were
expert systems, thinking instead, that they were only high level
interactive environments (to be fair Biesen classified these as first
generation expert systems). Third generation systems involve iconic
programming and the use of virtual instruments, best typified by Lab
View from National Instruments. The newest systems have large data bases
of instruments and also provide a feedback from/to the physical
instrument.

Almost all of the "fuzzy" papers were about applications, and presented 
by Japanese or Chinese. (I have already mentioned in several reports that 
the Japanese and Chinese are very aggressively applying these techniques 
to applications.) An exception was by 
	Laurent Foulloy 
	Laboratoire d'Automatique et de Micro-Informatique Industrielle
	University of Savoie
	B.P. 806, 74016 ANNECY Cedex, FRANCE
who gave an elegant description of sensors which compute and report
linguistic assessments of their values, a step beyond today's smart
sensors--Foulloy calls them symbolic sensors. This is basically
theoretical work; a symbolic sensor is a general component that can be
reconfigured to adapt itself to the measurement context. In the
discussion after this paper Foulloy was asked "why fuzzy?" Declining to
get into a long discussion, he observed that this was an easy way to
implement a nonlinear controller. (The more theoretical types in the
audience seemed unconvinced, but I thought this was an apt answer.)
Another non-application paper was by 
	Francois Terrier
	Dein/Sai Ce-Saclay, Fif/Yvette CEDEX F-91191 France
who discussed linguistic fuzziness.

Of the application papers about fuzzy logic, one worth mentioning was by 
	Toshio Fukuda
	Department of Mechanical Engineering
	Nagoya University
	Furo-cho, Chikusa-ku, Nagoya 464-01, JAPAN
who considered a system composed of multiple sensors (such as on a
robot) and discussed the problem of how to integrate data from
overlapping regions. His specific problem was using three distance
sensors, each accurate over different (overlapping) ranges. His
solution, use both a neural net and fuzzy logic.

At the far end, a few papers were about algorithms and problems that
would have been more appropriate at other conferences. But, I would
characterize the greatest fraction of the papers as being about
intelligent measuring techniques in the sense that some computing
capability was integrated with sensors for very specific problems such
as gravity flow, high precision temperature measurements, scanning
tunneling microscopy, capacitive tiltmeters, and others. These have
nothing to do with AI.  Several sessions specifically had "AI" in their
title, but I saw very little of that discipline in any of these papers.
However, as I have noted in past reports concerning descriptions of
experimental laboratory hardware, laboratory scientists exhibit a
tremendous amount of creativity with which techniques are adapted or
developed, and frankly I could not imagine how many of these solutions
would have been discovered without intimate collaboration with the
physical problem being studied.  In that sense I agree completely with
Zingales.

(3) Where were the US participants? TC7 is a European based
organization, and so one expects a preponderance of their scientists.
Nevertheless it was disappointing not to see other Americans at an
international English language meeting on such an important topic. I
wondered about this; the conference organizers did not have any ready
answers either.  They claimed that some related work is done in the US,
but it is more heavily based on electronics than the broader type
sensing of interest here.  IMEKO also publishes a journal,
"Measurement"; its editorial offices are in the UK. A scan through some
issues show papers mostly from Europe, but also from other places
such as Australia and Japan; there is essentially nothing from the US,
and I wonder if Americans read it. (Not surprisingly, the Australian
contributions are about intelligent measurement of coal dampers, etc.)

CONCLUSION.
This seems to be a topic ready to take off. There is an excellent match
between what we can measure and what one needs to compute based on these
measurements. AI and related computer science technology is going to be
adapted to help solve measurement problems. Not much has happened yet
compared to what will occur. Sensors are getting smart in that they can
perform substantial computation, but most of the "intelligence" is ad
hoc. There are a few theoretical developments that are being pursued,
but to me the field seems wide open and very attractive for fascinating
collaborative projects between hardware and software researchers.  The
modest amount of AI I saw was mostly being described by the Europeans;
with a few exceptions the Asian papers were much more application
oriented, albeit very clever.  At the moment it appears that the bulk of
the work will likely be done by the scientists crafting the experiments,
but people with more computer science training could learn a great deal
about real problems in addition to significantly contributing in an
important field.

LOCATION OF THE SYMPOSIUM.
The symposium was held at Ritsumeikan University, a private university in 
the north-east part of Kyoto, within walking distance of many famous 
sites. Ritsumeikan was established in 1895, but only in the past few 
years has begun to establish itself with graduate programs. Even so, it 
still has only a few hundred graduate students, although it has more than
20,000 undergraduates as well as a few thousand who go at night. Over
the years, Ritsumeikan has awarded almost 150,000 Bachelor's degrees,
1800 Masters, and 160 PhDs.  It now has active graduate programs in
Business Administration, Arts and Philosophy, and various aspects of
International studies. Graduate programs in Science and Engineering are
in place and growing. (Incidentally, it has recently been projected that
by the year 2005, Japan will be lacking 480,000 scientists and
engineers--300,000 in engineering, 130,000 in science, and 40,000 in
medical science/pharmacy, while it will have a 33,000 person oversupply
of liberal arts graduates.) Ritsumeikan has a 1.4 million volume
library, and a special depository of United Nations associated volumes,
numbering about 40,000 plus documents. For me, one of the interesting
aspects of the University is the degree to which it has become
international; there are exchange agreements with over a dozen large
universities in the US, Canada, UK, France, Germany, Poland, USSR,
China, and Australia.  Most of this is new, there are only about 300
non-Japanese students, mostly from Taiwan, Korea, China, and Malasia.
But this September about 100 Ritsumeikan students will spend an academic
year at the University of British Columbia, Canada. The campus is
attractive, with some (not all) new buildings (including the one that
the symposium was held at). Although at the moment there are very few
Western students, those looking for a Japanese experience should
consider it. For additional information, contact

	Dr. Masateru Ohnami
	President, Ritsumeikan University
	56-1 Kitamachi, Tojiin
	Kita-ku, Kyota 603 Japan
	 Tel: +81-75-465-1111

---------------------------------END OF REPORT----------------------------