Abstract
This dissertation presents two contributions which combine the domainsof data compression and evolutionary search. The first contribution is
the development of a new approach where genetic programming (GP) is used
to improve the compressibility of the data. A genetic programming system
is used to generate a transformation which reduces the entropy of the
data. Applying this transformation to the data, prior to its compression
by a standard compression algorithm, will allow it to be compressed to a
smaller size. The advantage of this approach is that it automates the
process where human design would otherwise be required. This
automatization allows us to generate a transformation for each
individual input file.
A second contribution of this work is a modification to a standard
genetic algorithm (GA) which introduces low level modularization. This
modularization is obtained by using a data compression technique to
shorten the length of the genotype, hence the name compressing GA
(c-GA). By compressing the genotype of the individuals we protect the
genetic material and make the reuse of combinations of features
possible. The validity of our assumptions has been tested with a series
of reference GA and GP problems. Results indicate that genotype
compression is beneficial to the search process and performs best for
problems which feature repetition in the solution representation.
Date of Award | 24 Nov 2006 |
---|---|
Original language | English |
Supervisor | Kris Steenhaut (Promotor) & Ann Nowe (Promotor) |
Keywords
- data compression