遺伝的プログラミングの非拘束な一次元表現

概要

論文の詳細を見る
In order to develop further the field of Genetic Programming(GP), it is necessary to experiment on different representations for programs manipulated by GP systems. Especially we think it is important to try linear representations, which enable to use more types of genetic operators and can be faster in evaluation and genetic manipulation of programs. The usual representation of programs in GP is a function. tree, but there also have been some researches in linear representations. Keith & Martin proposed a linear representation but their genetic operators are syntactically constrained so that their representation is equivalent to the tree representation. Perkis uses a linear representation with a syntactically unconstrained crossover but his functions are limited to arithmetical functions; he does not use instructions like if, and he can not evolve complex programs. Banzhaf uses the if instruction but his representation requires to repair the programs after genetic recombination which is equivalent to folding the search space and leads usually to poor performance. We propose a linear and syntactically unconstrained representation which does not require repairing and enables to use the if instruction.
社団法人電子情報通信学会の論文
1995-09-05