agent-based simulation systems (e.g. Swarm, SIMEX)
If you have suggestions for workshops you would
like to see us hold, please send email to
cscslab@umich.edu.
Upcoming workshops: The following workshops are
currently scheduled.
These workshops are open to all, but space is limited.
If you are interested in attending, please send a message to
cscslab@umich.edu
to register.
If you are interested in one of these workshops,
but cannot make the day/time they are scheduled for,
please let us know---if enough people are interested,
additional workshops will be scheduled later in this term.
Title: Evolutionary Algorithms -- An Introductory Survey
Date: Tuesday 7 November 2000
Time: 3:10-5:00pm
Place: Room 4246 Randall Bldg.
Evolutionary Algorithms (EAs) are general-purpose adaptive algorithms
which use the basic ideas of natural evolutionary processes. A number
of different EAs have been studied, but they all share some common
elments and processes. In particular, each evolutionary algorithm
starts with a population of structures (individuals), often randomly
generated. These structures are evaluated (assigned a fitness) based
on their performance in an environment. The better structures are
allowed to reproduce and the structures which do not perform as well
are eliminated. Variation is introduced into the offspring
structures, providing a mechanism for exploration of the space of
possible structures. Many variants perform no better or even worse
than the parents, but some offspring perform better. This
generational process is repeated, leading to continuing adaptation of
the structures.
By defining the appropriate structures and evaluation function, EAs
have been used in a wide variety of search and optimization problems,
as well as in many different kinds of machine learning systems. EAs
also are commonly used as the main adaptive mechanism in models of
complex adaptive systems in the biological and social sciences,
including most of the systems sometimes referred to as 'artificial
life" and "artificial societies".
This talk will introduce the three main classes of EAs, including
Genetic Algorithms (GA), Evolutionary Strategies (ES), and
Evolutionary Programming (EP), as well as two more recent variants,
Genetic Programming (GP) and Evolution Programs. Examples of each EA
will be discussed, and the differences between the approaches will be
outlined. Handouts will include references to sources of further
information, including software packages available in the CSCS Lab and
via the Internet.
The workshop is free, but registration may be limited. If you are
interested in attending the workshop, please let us know by sending
email to cscslab@umich.edu.
Back to Top of this Page
Title: Introduction to Genetic Programming
Date: Tuesday 28 November 2000
Time: 3:10-5:00pm
Place: Room 4246 Randall Bldg.
Genetic programming (GP) uses the ideas of natural selection to create
computer programs that solve user-specified problems. That is, an
initial population of randomly generated computer programs are
evaluated to rate how well they solve the problem at hand. The better
programs are allowed to reproduce and the programs which do not solve
the problem as well are eliminated. Most importantly, variation is
introduced into the offspring programs by allowing their parent
programs to "mate" (i.e., exchange subprograms). Many variants are
worse than the parents, but a surprising number are also better at
solving the problem. This generational process is repeated, with the
best programs at each generation becoming better and better at solving
the problem, until the user is satisfied (or runs out of computer
time!).
Surprisingly, genetic prgramming has been shown to solve a wide
variety of standard problems encountered in the machine learning
literature. GP has been applied to many different kinds problems,
including pattern recognition and classification, robot control,
neural net design and learning, induction and regression problems, and
even the creation of art. Most recently there is much effort applying
GP to "real world" problems.
This workshop will begin with an introduction to the basic ideas of
GP, including how problems and the programs that solve them are
represented, and how the underlying genetic algorithm works to select
and recombine programs. Several public domain software packages which
implement genetic programming will be described. An example or two
will be examined using one of these GP packages on the Lab computers.
Students will be able to use the any of the packages to define and
solve their own problems using GP.
The workshop is free, but registration is limited. If you are
interested in attending the workshop, please let us know by sending
email to cscslab@umich.edu.
Back to Top of this Page
