Population Structures and Genetic Algorithms
Mark Smucker
Department of Computer Sciences
University of Wisconsin-Madison
smucker@cs.wisc.edu
2:30 pm Fri. Mar. 31 in 2310 Computer Sciences and Statistics Bldg.
Genetic algorithms are one variant of evolutionary search
techniques that operate via selection, recombination, and mutation of
a population of potential solutions to a problem. One problem faced
by genetic algorithms is ``premature convergence'' in which the
population becomes genetically homogeneous before a good solution has
been found. Many researchers have reported that giving the genetic
algorithm a population structure which restricts mating can reduce the
problem of early convergence and even in some cases find good
solutions faster than the regular genetic algorithm. This talk will
review genetic algorithms and the ways population structure has been
used by other researchers and will also present original work being
done on trying to understand the properties of various population
structures. No prior knowledge of genetic algorithms will be needed
to understand this talk. (This work is in collaboration with Dan
Ashlock of Iowa State University.)