| Basics, September 13, 1999, 10:00 - 12:00 h
A: Fuzzy Logic
Hans-Jürgen Zimmermann, ELITE-Foundation, Germany
Fuzzy Set Theory started in 1965 as a formal theory but with
the motivation to help improve problem solving and model human reasoning and
decision making. In the meantime there exist more than 25,000 publications in
this area, the theory has been extended considerably, it has partly been
verified empirically and it has been applied successfully to many real
problems.
For a newcomer who is interested in one of the application
areas covered by the other tutorials it will be difficult to find the best
entry into the theory and its basic application areas. It is the intention of
this tutorial A to bridge this gap. An introduction will be given into basic
fuzzy set theory. Different types of fuzzy sets will be discussed and also
extensions of the basic theory by families of operators (t-Norms, t-conorms,
averaging operators). The basic terminology for fuzzy decision making will be
explained and the audience will be introduced to the most important application
areas, such as fuzzy control, intelligent data analysis, approximate reasoning,
fuzzy expert systems and fuzzy multicriteria analysis. Specific areas or
questions in which the participants of this tutorial are interested can also be
discussed.
B: Neural Networks
Mark Plumbley, Kings College London, United Kingdom
Neural networks are computational models, inspired by the
operation of networks of neurons in the brain, which can learn to perform
useful classification and analysis tasks on data. This tutorial will give an
introduction to the principles of neural networks, see what types of tasks
neural networks are suitable for, and consider how to apply neural networks to
new problems.
We shall concentrate on probably the two most popular types
of neural network: the Multilayer Perceptron using Error Back Propagation
("Backprop") and the Kohonen Self-Organizing Map (SOM). We will
concentrate on visual interpretation of the learning principles and algorithms,
avoiding the use of complex mathematics. We will explore how to ensure that the
trained network "generalizes" well to new data, and adaptations to
the standard networks to help them learn time-related information and to learn
more quickly and reliably.
We will show how the development of systems to tackle new
tasks using neural networks is different to normal system development. We will
cover the special considerations required for neural network projects, such as
data collection and preparation and prototyping using the neural network
itself.
Finally, an indication of current and future developments of
neural networks will be given, showing how they are developing links with other
fields. Finally, pointers to sources of further information and assistance will
be given (e.g. books, WWW pages, societies, etc.) to allow participants to find
out more in their own area of interest.
C: Genetic Algorithms
Terry Fogarty, Napier University of Edinburgh, UK
An introduction to evolutionary computation including a
detailed description of a simple evolutionary algorithm and a broad
introduction to genetic algorithms, evolutionary programming, evolution
strategies and genetic programming. This will be followed by some descriptions
of particular applications of evolutionary computing in control, classification
and communications.
D: Knowledge Discovery in
Databases
Rüdiger Wirth, Daimler Benz AG, Germany
Data Mining or Knowledge Discovery in Databases is the
process of identifying valid, novel, potentially useful, and ultimately
understandable patterns in data. It is a rapidly evolving new interdisciplinary
field based on applied statistics, databases, pattern recognition,
visualisation, and artificial intelligence.
This tutorial will give a broad overview of the main issues
of knowledge discovery in databases. We will focus on the data mining process
and present a comprehensive process model. Various techniques for different
phases of the data mining process - such as data preparation, modelling and
evaluation will be covered.
Case studies from various areas will illustrate the
practical applicability and the business potential of data mining.
|
