Аннотация
Abstract This paper presents an empirical study of the
variation of program size over time, for a form of
Genetic Programming called Cartesian Genetic
Programming. Two main types of Cartesian genetic
programming are examined: one uses a fully connected
graph, with no redundant nodes, while the other allows
partial connectedness and has redundant nodes. Studies
are reported here for fitness based search and for a
flat fitness landscape. The variation of program size
with generation does not behave in a similar way to
that reported in other studies on standard Genetic
Programming. Depending on the form of Cartesian genetic
programming, it is found that there is either very weak
program bloat or zero bloat. It is argued that an
important factor in the analysis of the change of
program length is neutral drift, and that if genotype
redundancy is present, the genetic neutral drift
simultaneously improves search and compresses program
code.
Пользователи данного ресурса
Пожалуйста,
войдите в систему, чтобы принять участие в дискуссии (добавить собственные рецензию, или комментарий)