Genetic algorithms with implicit memory
Date
2011
Authors
Morris, Robert
Journal Title
Journal ISSN
ISSN
DOI
Volume Title
Publisher
De Montfort University
Peer reviewed
Abstract
This thesis investigates the workings of genetic algorithms in
dynamic optimisation problems where fitness landscapes materialise
that are identical to, or resemble in some way, landscapes
previously encountered. The objective is to appraise the
performances of the various approaches offered by the GAs.
Approaches specifically tailored for different kinds of dynamic
environment lie outside the remit of the thesis.
The main topics that are explored are: genetic redundancy,
modularity, neutral evolution, explicit memory, and implicit memory.
It is in the matter of implicit memory that the thesis makes the
majority of its novel contributions. It is demonstrated via
experimental analysis that the pre-existing techniques are
deficient, and a new algorithm – the pointer genetic algorithm
(pGA) – is expounded and assessed in an attempt to offer an
improvement. It is shown that though it outperforms its rivals, it
cannot attain the performance levels of an explicit memory algorithm
(that is, an algorithm using an external memory bank).
The main claims of the thesis are that with regard to memory, the
pre-existing implicit-memory algorithms are deficient, the new
pointer GA is superior, and that because all of the implicit
approaches are inferior to explicit approaches, it is explicit
approaches that should be used in real-world problem solving.