A Genetic Algorithms Workshop
Summer Programs at MIT
Cambridge, MA, USA
17-21 June 1996
Since their conception in the 1970's, genetic algorithms (GAs) have
stimulated theoretical interest in evolutionary processes and their
simulation. Practical applications have also been investigated since
that time, but have generated a large amount of interest more recently.
The robustness and versatility of GAs have motivated their application
to a diverse range of problems, such as scheduling, structural
optimization, time series forecasting, computer animation, image
processing/recognition, and software generation. As computation becomes
increasingly more powerful and inexpensive, more applications will
This course is intended for anyone who is interested in any type of
optimization. This includes engineers, managers, economists, computer
scientists, and those interested in planning, operations, and
operations research. There are no limitations with regard to particular
technical disciplines or industries. We also welcome those more
oriented toward basic sciences, such as biology and sociology, where
evolutionary models may be useful.
The course is aimed at the practitioner who will develop programs that
use evolutionary methods. For practical purposes, genetic algorithms
are inherently simple in concept and application. As such, the course
will move participants quickly to actually use GAs on real problems.
Existing theory will be presented to the extent needed to understand
the operation and performance of GAs and to describe their historical
context. Participants should bring to the course a real problem from
their own endeavors that they wish to solve.
Instead of being derived from a particular mathematical model, genetic
algorithms are formulated to provide a simplified simulation of natural
evolution. Designs (more formally, points in a design space) are
to organisms involved in a process of natural selection. Designs exist
generations where they mate, reproduce, mutate, and die. Optimization
performed as the fittest (the best with respect to our optimization
designs in a generation are chosen to serve as parents for the next
generation. Parents mate and create children that possess a combination
their characteristics. This is achieved by recombination of a
representation of the designs, just as in nature. With increasing
generations, the average fitness of a population increases as does the
fitness of the best individual.
Because GAs do not require derivatives (or other extra information) and
because they sample the design optimization objective throughout the
space, they are usually very robust and typically perform well in
ill-behaved search spaces. An additional advantage of GAs is the
chromosome representation, which allows the easy encoding of almost any
type of design space and the subsequent evolutionary optimization.
Course participants are encouraged to bring optimization problems they
currently face. A significant portion of the workshop is dedicated to
developing working optimizations and models of real problems. Unix
workstations will be provided for examples and visualization exercises,
but participants may use MacOS or PC variants as well.
For more details and registration information, see
For complete information about *all* of MIT's summer programs, see
For more details about GAlib (a C++ genetic algorithms library), see
Please direct inquiries regarding course specifics to jakiela at mit.edu
Please direct inquiries regarding GAlib to mbwall at mit.edu