FLAI'93 CONFERENCE REPORT

Fuzzy Logic in Artificial Intelligence
8th Austrian Artificial Intelligence Conference, FLAI'93
Linz, Austria, June 28-30, 1993

Gabi Thalhammer
TU-Wien, Information Systems Department, Paniglgasse 16, 
A-1040 Wien, Austria, Europe. 
Email: thal@vexpert.dbai.tuwien.ac.at

Overview 

The Eighth Austrian Artificial Intelligence Conference, which took
place at the Bildungszentrum St. Magdalena, Linz, Austria, June 28-30,
1993, was devoted to the subject of "Fuzzy Logic in Artificial
Intelligence" to acknowledge the increasing importance of fuzzy logic
in many areas of applications. The conference hosted by the Austrian
Society for Artificial Intelligence was organized by Erich Peter
Klement from the Fuzzy Logic Laboratorium Linz, Johannes Kepler
University, Austria, and Wolfgang Slany from the Christian Doppler
Laboratory for Expert Systems, Technical University Vienna, Austria.
Klement has been contributing to the early development of fuzzy set
theory in Austria for a long time, after Lotfi A. Zadeh had published
his first paper on "Fuzzy Sets". The conference aimed at introducing
the subject of theoretical fuzzy logic and its applications to members
of the industry as well as updating researchers in the field of fuzzy
logic and AI. Tutorials, workshops and talks covering theoretical
issues, machine learning, expert systems, robotics & control,
applications to medicine, and applications to car driving were held.
During the breaks a hard- and software exhibition could be visited
where several companies and university laboratories presented their
work.  OMRON (Japan) presented their compact digital fuzzy processors
for control supported by a full line of development tools.
Additionally, a comprehensive book exhibition was offered. 88
participants from 15 countries attended this international conference.

The conference featured invited talks by Lotfi A. Zadeh from the
Computer Science Division and The Electronics Research Laboratory of
the University of California, USA, and Irina V. Ezhkova from the
Russian Academy of Sciences, Moscow, Russia, and both talks were very
much appreciated by the audience. Zadeh pointed out that we all have
witnessed a significant increase in the rate of growth of MIQ (Machine
Intelligence Quotient) of consumer products and industrial systems.
One reason, he states, is the rapidly growing use of soft computing
and especially fuzzy logic in the conception and design of intelligent
systems. Ezhkova presented her work on context formalization based on
fuzzy set theory which led to the development of contextual systems
for flexible decision-making in fuzzy environments. She provided some
results taken from an application for investigation of Sudden Infant
Death Syndrome. A special keynote address on "Neural Nets to the Aid
of Steelmaking Industries" was given by Brahma Deo from the Indian
Institute of Technology, Kanpur, India.

Tutorials 

The Tutorials were given in German to especially address the
participants coming from the industry. Members of e.g. VOEST-ALPINE
Stahl, OMRON Electronics, VMV AG, Siemens AG, BEKO GesmbH, and many
researchers, attended tutorials to get a deeper understanding of fuzzy
logic and a look at recent scientific results. Hans-Juergen Zimmermann
from RWTH/ELITE, Aachen, Germany, introduced the basics and methods of
fuzzy set theory, emphasising on the field of fuzzy inferencing.
According to Zimmermann, the methods of fuzzy set theory, which have
been developed over the past 25 years, are to a great extent the basis
for the area of expert systems. Applications such as strategic
planning and the control of flexible production systems, which use
fuzzy inference, were discussed. Rudolf Kruse from the Technical
University Braunschweig, Germany, compared classical PID control
methods to those of fuzzy control.

Workshops

Workshop for Doctoral Students in Fuzzy-Based Systems: 
Johann Gamper from the Department for Medical Computer Sciences,
University of Vienna, Austria, and Bernhard Moser from the Institute
for Computer Sciences, University of Salzburg, Austria, encouraged 8
Ph.D. students from European universities, who are working in the
field of fuzzy-based systems, to meet, discuss their research and
develop their working skills together with some senior researchers in
a one day workshop. Slany introduced the fuzzy-mail list and related
services.  He stressed the possibility of a wide access to
information, resources and methods. Zimmermann gave some hints on "How
to write a Ph.D.-thesis".  Several senior researchers contributed
their experience, so finally the subject was extended to "Writing
Ph.D.-thesis in different countries". Ludmila Kuncheva from the
Bulgarian Academy of Sciences in Sofia, Petr Hajek from the Academy of
Sciences in Prague, and Anio O. Arigoni from the University of Bologna
revealed the current situations in their countries.

Workshop on Fuzzy Logic for Commercial and Industrial
Applications:
Andreas Geyer-Schulz from the Institute for Information Systems,
Business University Vienna, Austria, and Peter Kotaucek from BEKO
GesmbH, Vienna, Austria, discussed with 15 participants from 7
countries the management and design decisions necessary for building
fuzzy applications along with the question whether fuzzy logic should
be introduced already at high-school level.  Furthermore Arkadiusz
Salski from the University of Kiel, Germany, presented a paper on
"Fuzzy-Cluster Analysis and Fuzzy-Modelling in Ecological Systems",
and Monika Reiter from Joanneum Research, Graz, Austria, talked about
"Expert Systems and Fuzzy Logic Projects and Experiences in Austria".
A tool for designing control systems called "FS-Fuzzy Soft Operating
System" from Fuzzy Soft AG, Switzerland, was introduced by Thomas
Tilli from Development Research, Aachen, Germany. Concerning the
question of the length of fuzzy inference chains, Tilli suggested
defuzzification after each step.

Workshop on Fuzzy Scheduling Systems:
The workshop opened with a summary by Roger Kerr from the Technical
University of Vienna, Austria, of the major research issues and
challenges in Fuzzy Scheduling. The main application was identified as
in the representation, propagation and relaxation of imprecise
constraints and in the evaluation of schedules with respect to vague
goals. Important issues were felt to be the exploitation of fuzzy
constraint representation to find new and more efficient ways of
factoring the search space. Luis Custodio and Carlos Pinto-Ferreira
from the Technical University of Lisbon, Portugal, presented a
hierarchical production management system consisting of a fuzzy
decision maker, a fuzzy controller and a fuzzy scheduler. All three
levels make use of linguistic variables to specify different types of
control decisions which are fuzzily matched to the situation
pertaining. Helene Fargier from the Universite Paul Sabatier,
Toulouse, France, presented her work on an extension of the OPAL
scheduling framework to the handling of fuzzy temporal constraint
propagation induced by user specification of fuzzy release and due
dates and processing times. Her experiments showed that fuzzy analysis
is more productive than crisp analysis without an increase in
computational expense. Finally, Slany presented a fuzzy constraint
relaxation technique applied in conjunction with an Expert Scheduling
System for use in the steelmaking industry. The main fuzzy constraints
in this system were related to chemical compatibility constraints
between successive jobs.

Workshop on Karl Menger, Fuzzy Logic and Artificial
Intelligence - An Experiment in Reflection:
Rainer Born from the Institute for Philosophy and Philosophy of
Science presented an unorthodox "investigation" into the logical and
philosophical foundations as well as the historical developments
concerning the relationship between fuzzy logic and artificial
intelligence. Born put special emphasis upon the mathematical and
philosophical contributions of Karl Menger. Klement took up this idea
and showed how some of Karl Menger's work on probabilistic metrics, in
a systematical context, are essential for the development of fuzzy set
theory, especially with respect to the concept of "t-norms".  Avi
Akkermann from the University of Saskatchewan, Saskotoon, Canada,
introduced his work on "Demographic Measurements" using Karl Mengers
concept of "fluents", showing the importance of "fluents" and the
tight connection to fuzzy set theory. Born presented the idea of
concept-formation (due to fuzzy set theory or other approaches - e.g.
connectionist ones) and the way in which our "outer world" is grasped
or represented.  According to his general scheme for connecting
"language-information and reality", there are some concepts which
provide the basis for sound information processing by human beings,
giving credit to both the scientific and common-sense approaches of
knowledge acquisition.

Technical Program

Theoretical Issues:
Anio O. Arigoni from the University of Bologna gave an introduction of
the "Typicality of Concept Instances: a Semiotic Way for its
Evaluation" which seemed to have been of more interest to cognitive
psychologists as it was addressed to proving that fuzzy set theory
does not provide the essential features for conceptual modeling. His
viewpoint was strongly opposed by Klement and other participants who
have already proved the efficiency of the concept of fuzzy models.
Petr Hajek from the Academy of Sciences Prague, Czech Republic,
introduced a comparative fuzzy modal logic by investigating a formal
logical system dealing with both uncertainty (possibility) and
vagueness (fuzziness). The system is many-valued and modal, and also
related to a many-valued tense logic. Hajek then presented the proof
of completeness of his idea, which was very well received.

Machine Learning:
Detlef Nauck from the Technical University Braunschweig, Germany,
discussed four possible combinations of neural networks and fuzzy
controllers to tune membership functions.  To enhance Berenji's ARIC
architecture (a neural network for tuning fuzzy logic controllers
through reinforcements) a neural fuzzy controller that uses a learning
algorithm, based on a fuzzy error measure, was suggested. Ann Nowe
from the University of Brussels presented a reinforcement learning
algorithm for obtaining a robust controller which locates the "safest"
regions within the viability domain to drive and keep the system in a
safe state. A new method for designing intelligent tutoring systems
using genetic algorithms (GAITS) was introduced by Mohamed Quafafou
from the Technical University of Nantes, France. Due to the complexity
of describing pedagogical concepts, a method combining fuzzy logic and
genetic algorithms was attempted. He gave some preliminary results of
the system, computing an optimal strategy to evolve a bad and a good
student's knowledge from an initial profile.  To round up this
session, Hideyuki Takagi from the University of California, Berkeley,
USA, emhasized the importance of reducing the development time and
increasing the performance of fuzzy systems. His approach combines the
advantages of neural networks and genetic algorithms for fuzzy system
design.

Expert Systems:
Mario Fedrizzi from the University of Trento, Italy proposed an
approach to consensus reaching based on linguistically expressed
individual opinions and on opinion changing aversions, explaining
some. Heinrich J. Rommelfanger from the J. W. Goethe-University
Frankfurt am Main, Germany investigated the suitability of methods
used in fuzzy control for building non-technical expert systems, and
demonstrated an expert system to determine creditworthiness.  Applying
linguistic variables and employing fuzzy conjuction methods enables
engineers to capture the human reflection process more appropriately
and thus to construct expert systems that might deserve that name.

Robotics & Control:
Bernhard Bloechl from Germany described how real time path tracking is
performed by fuzzy control. For vehicles that only follow a path,
without consideration of obstacles, a fuzzy controller was
successfully developed and tested on ATHENE (a three-wheeled vehicle
with front-wheel drive). The controller, a path finder (single
behaviour) module, consisting of two cooperating submodules: an
image-processing module to extract path information with a CCD camera
and a fuzzy-control module to control the actuators of the vehicle,
was roughly outlined.  Vahe K. Yeghiazarians from the Technical
University of Vienna, Austria, demonstrated a fuzzy rule based robot
controller. By progressively adding displacement increments,
determined by fuzzy control, the robot arm moves. He showed that by
using a fuzzy control loop, the calculation of the inverse kinematics
of a robotic manipulator can be bypassed, which will lead to several
advantages (e.g. treating obstacle avoidance problems by means of
linguistic variables). A very demanding presentation was given by
Harri Koskinen from the Technical Research Centre of Finland, Tampere.
He investigated active magnetic bearing (AMB) systems which, due to
their nonlinear structure, can be effectively controlled by fuzzy
controllers.  The principle of AMB is the following: A rotor hovers
between two magnets due to the attractive force of the electric
magnets. AMB systems are used for instance as energy storage systems
in satellites. A displacement sensor measures the rotor deviation from
the reference position and this measurement is used by the controller.
The control output determines the currents that the power amplifier
supplies to the magnets. Koskinen outlined a fuzzy derivate gain
adjusting method for AMB applications. He showed that with his method
he can decrease the effects of the vibration noise.

Applications to Medicine:
Silvio Giove from the University of Trento, Italy, gave a description
of a fuzzy adaptive control technique to automate the dialysis
procedure. Ludmila I. Kuncheva presented an idea about a
classification scheme combining tools from AI with fuzzy pattern
recognition, trying to simulate a physician's decision process. An
example from the aviation medicine demonstrated a higher
classification accuracy than obtained with linear discriminant
analysis. Furthermore, she investigated an application of the scheme
to the medical diagnosis expert system CADIAG-2.

Applications to Car Driving:
Friedhelm Muendemann, FH Brandenburg, Germany, presented some positive
results from a test system using fuzzy concepts for predicting the
behaviour of other drivers on a highway. The module shall be
integrated in a multi-transputer environment and evaluated on
short-distance runs. Alexandra Weidmann from FU Berlin, Germany, uses
fuzzy controllers to determine and control the distance between two
moving cars.

Highlights

Zadeh certainly inspired the participants to keep on doing their best
in developing, applying and presenting fuzzy logic concepts. The
panels on FL & AI made the participants take part in engaged
discussions on future expectations. Slany gave a very amusing
finishing talk. He compared the European slow adoption of new and
revolutionary ideas, like fuzzy logic, to the much faster way the
Japanese adopt such ideas. He also stressed the importance of
encouraging innovativeness and of active exchange of know-how.

Conclusion

The conference was very well organized. Its pleasant atmosphere, with
lots of chatting time between sessions, a secluded conference
location, as well as excellent weather conditions, helped bringing
together experienced researchers and members of the industry
interested in practical applications of fuzzy logic. As outlined, many
different ways of applying fuzzy logic have been introduced and
discussed at this conference.  Concepts of other AI disciplines like
robotics, expert systems, diagnosis, learning, modal logics, and
belief networks have been successfully combined with fuzzy logic to
find new and better solutions to AI problems.

Proceedings

E. P. Klement, W. Slany: "Fuzzy Logic in Artificial
Intelligence", Proceedings of the 8th Austrian Artificial
Intelligence Conference, FLAI '93. Lecture Notes in Artificial
Intelligence 695, Springer-Verlag 1993.

-----------------------------------------------------------------
Erich Peter KLEMENT
FLLL-Fuzzy Logic Laboratorium Linz, Institut fuer Mathematik
Johannes Kepler Universitaet, A-4040 LINZ, Austria, Europe.
Email: k312570@edvz.uni-linz.ac.at
Phone (++43) (732) 2468-9151, Fax (++43) (732) 2468-10
-----------------------------------------------------------------
Wolfgang Slany
Christian Doppler Laboratory for Expert Systems, Information
Systems Department, E 184/2,  Paniglgasse 16, A-1040 Wien,
Austria, Europe. Email: wsi@vexpert.dbai.tuwien.ac.at
Phone (++43) (1) 58801-6123, Fax (++43) (1) 5055304
-----------------------------------------------------------------
Send GET FUZZY-MAIL INFO to listserver@vexpert.dbai.tuwien.ac.at
for information about the fuzzy-mail list.
-----------------------------------------------------------------