A New GP Recombination Method Using Random Tree Sampling

Tanji, Makoto; Iba, Hitoshi

doi:10.1541/ieejeiss.130.775

Cited by 3 publications

(2 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In order to improve the performance of GP, various methods have been proposed: the method generating individuals in the next generation by joining fragments of the tree structure that randomly been sampled from several parent individuals [5], the method extracting useful tree structures from individuals called frequent trees [6], [7] that are subtrees that frequently appear in the population, the island model that combines those frequent trees [8], the method using the Semantic Aware Crossover (SAC) [9] that uses the similarity of subtrees to avoid destructive of tree structures, and the method in which semantics are used for select operation to keep diversity [10].…”

Section: Introductionmentioning

confidence: 99%

Extension of Genetic Programming with Multiple Trees for Agent Learning

Igarashi¹,

Takahashi²,

Inaba³

2016

JCP

View full text Add to dashboard Cite

This paper proposes an extension of genetic programming (GP) with multiple trees. In order to improve the performance, GP with control node (GPCN) and its three kinds of modification have been proposed. In GPCN, an individual consists of several trees which have the number P of executions. In previous work, the two kinds of modification, the conditional probability and the cross-cultural island model are employed. This paper proposes two methods: the new island model that combines the conditional probability with two islands in the cross-cultural island model and a method exchanges multiple trees in an individual in a suitable order. Experiments are conducted to show the performance in the garbage collection problem and the Santa Fe Trail problem.

show abstract

Section: Introductionmentioning

confidence: 99%

Extension of Genetic Programming with Multiple Trees for Agent Learning

Igarashi¹,

Takahashi²,

Inaba³

2016

JCP

View full text Add to dashboard Cite

show abstract

“…Evolutionary methods are known to be able to obtain optimum rules for agent action in a broad search space. Among evolutionary methods, genetic programming (GP) and genetic network programming (GNP) have been investigated eagerly and widely (Koza, 1992;Hirasawa et al, 2001;Iba, 2002;Mesot et al, 2002;Tanji and Iba, 2010). Genetic network programming (GNP) is also known to able to find better solutions than genetic programming (GP) can (Hirasawa et al, 2001;Iba, 2002).…”

Section: Introductionmentioning

confidence: 99%

Experiments Assessing Learning of Agent Behavior using Genetic Programming with Multiple Trees

Igarashi

Takahashi

Inaba

2014

Proceedings of the 6th International Conference on Agents and Artificial Intelligence

View full text Add to dashboard Cite

In this paper, experiments to assess agent behavior learning are conducted to demonstrate the performance of genetic programming (GP) with multiple trees. Using the methods, each has a chromosome representing agent behavior as several trees. We have proposed two variants using the conditional probability and the island model to improve the methods' performance. In GP using the conditional probability, individuals with high fitness values are used to produce conditional probability tables to generate individuals in the next generation. In GP using the island model, the population is divided into two islands of individuals: one island maintains diversity of individuals. The other emphasizes the accuracy of the solution. Moreover, this paper improves methods to seek the optimal number of executions of each tree in an individual. Those methods are applied to a garbage collection problem and a Santa Fe Trail problem. They are compared with traditional GP, GP with control nodes, and genetic network programming (GNP) with control nodes. Experimental results show that our methods are effective for improving the fitness.

show abstract