 |
Research Interests
My research focuses on the class of evolutionary algorithms, algorithms
whose problem-solving ability is modeled after biological evolution. These
algorithms essentially operate via incremental solution optimization,
and have been show effective on a wide variety of problems without requiring
knowledge of a problem's domain. The ability of these algorithms to operate
without explicit domain knowledge makes them extremely useful in handling
problems whose domain is either unknown or inconvenient to model. Presently,
I am studying two relatively new algorithms, EuA and TEAM (described below),
in an effort to understand their behavior and, eventually, improve upon
their capabilities.
|
|
|
Projects
|
 |
|
EuA: The Eugenic Algorithm (cs.utexas.edu/users/malden/EuA/)
EuA (Prior, 1998) is similar to the Genetic Algorithm, however,
new chromosomes are generated via a function of an entire population
rather than from the recombination of, typically, two other chromosomes.
One assumes that an entire population contains more information
than just a few of its members, and should therefore allow "smarter"
construction of chromosomes. EuA constructs new chromosomes by statistically
analyzing gene/allele combinations to identify those which have
the greatest effect on chromosome fitness, and then incorporating
these combinations into the new chromosome. EuA has been shown to
perform quite well against other optimization techniques on a variety
problems.
|
|
TEAM: The Eugenic Algorithm with Modeling (cs.utexas.edu/users/malden/TEAM/)
TEAM (Kesteren, 2000) is an improvement upon EuA. Again, new chromosomes
are constructed from the analysis of an entire population, however,
TEAM modifies this analysis in two important ways. First, TEAM builds
and maintains a model of gene interdependencies, allowing more sophisticated
gene/allele analysis. Second, the analysis focuses on finding genes
which have a preferred allele, meaning an allele commonly appearing
in high fitness chromosomes. These improvements to EuA allow TEAM
to perform significantly better than EuA (and other optimization
techniques) on some problems.
|
|
|
