Views
3 years ago

Recombination Strategy Adaptation via Evolution of Gene Linkage

Recombination Strategy Adaptation via Evolution of Gene Linkage

Recombination Strategy Adaptation via Evolution of Gene

Recombination Strategy Adaptation viaEvolution of Gene LinkageJim SmithFaculty of Computer Studies & MathematicsUniversity of the West of EnglandBristol, BS16 1QYU.K.jim@ics.uwe.ac.ukAbstractThis paper analyses recombination strategies resultingfrom evolving gene linkage on a variety of fitness landscapeswith known properties. The evolution of “blocks” of linkedgenes within a genepool makes it possible for the recombinationstrategy to vary in both the amount of genetic materialexchanged during the formation of a new individual, and inthe number of parents contributing genetic material. The strategiesevolved are examined in the light of the known propertiesof the landscapes.The results obtained explain the previously recorded goodperformance of the algorithm as a self-adaptive mechanismfor function optimisation, as the amount and type of recombinationis able to adapt to suit the landscape being searchedunlike conventional “fixed” operators.1: IntroductionIn recent years the interaction between the Genetic Algorithm(GA) and Evolutionary Strategies communities has ledto an increasing interest in the use of adaptive operatorswithin genetic algorithms e.g. [1, 2, 3], as it has been recognisedthat the optimal rate of application of a given operatorwill depend both on the nature of the landscape beingsearched and on the current state of the search.The aim of this work is to develop a representation andoperators which allow for self-adaptation of the genetic searchto the nature of the landscape. This is intended to provide arobust function optimisation method which is not prone to thepoor performance that “fixed” operators can show on certaintypes of landscape.This paper examines a model within which the evolutionof heritable “links” between neighbouring genes provides thebasic means for the self adaptation of recombination strategies.Measurement of the amount of gene linkage also providesa natural means of analysing the recombinationbehaviour of the algorithm as evolutionary time progresses.This algorithm has already been shown to be competitiveagainst a variety of other recombination strategies and acrossa range of mutation rates when tested in the setting of a generationalGA [4]. Subsequent experiments have shown that thisT.C. FogartyFaculty of Computer Studies & MathematicsUniversity of the West of EnglandBristol, BS16 1QYU.K.tcf@ics.uwe.ac.ukpotentially disruptive mechanism works better in a “steadystate” setting, a result which was also found in work on multiparentrecombination techniques [5]. This paper thereforeconcentrates on analysing how the nature of the search adaptsto different types of landscapes.Kauffman’s NK model is used to provide a means of creatingand characterising the landscapes searched in terms of theamount of interaction between loci (epistasis). This is brieflydescribed in Section 3.2: The LEGO Model for Adaptive Evolution ofGene LinkageThe encoding and recombination used for this work, anddescribed below, is the “LEGO” model (Linkage EvolvingGenetic Operator) [4]. The recombination operator works byconsidering the GA’s population as a gene pool comprised of“blocks” of genes defined over certain loci, with these blockspotentially varying in size from a single gene to an entirechromosome. In order to achieve this each gene has associatedwith it two boolean flags determining whether it will link tothe genes to its left and right, and two adjacent genes are saidto be linked if the appropriate flags are both set to “true”.Blocks are thus comprised of chains of linked genes. This representationis similar to Schaffer and Morishima’s use of“punctuation marks” to encode for crossover points [1], howeverin this case the recombination is not restricted to two parents.A new individual is created by holding a series of competitionsto fill its loci, starting at the left hand edge. When agiven position is filled a choice is made from all the blockseligible to fill that space (i.e. whose left-most defined positionis at the locus being filled) based on their fitnesses. The entireblock (with its links) will be copied. A new competition isthen held to fill the next locus after the end of the block. Sinceeach block comes from a complete parent, it will always bepossible to find at least one suitable candidate block for eachcompetition. This is shown schematically in Figure 1.If all of the blocks are of unit length (no links) then thisreduces to considering the problem as a set of separate populationsof alleles (one for each locus) which is the basis of BitSimulated Crossover [6]. At the other extreme, if a chromo-

aDaptation strategies to spatial planning anD regional Development
A Gradient Oriented Recombination Scheme for Evolution Strategies
Adaptive Strategy Selection in Differential Evolution
Enhanced Differential Evolution with Adaptive Strategies for ...
Recombination and the evolution of mutational robustness
Self Adaptive Recombination and Mutation in a Genetic Algorithm
The Evolution of Sex and Recombination in Response to ... - Genetics
adaptive reptile color variation and the evolution of the mc1r gene
Linkage Strategies for Mapping Genes for Complex Traits in Man
An Efficient Strategy for Gene Mapping Using Multipoint Linkage ...
ADAPTIVE MOLECULAR EVOLUTION IN THE OPSIN GENES OF ...
Analysis of recombinative algorithms on a non-separable building ...
Adaptive Strategy Selection in Differential Evolution - HAL - INRIA
Cycling Co-Evolution Resulting from Genetic Adaptation in Two ...
The effect of recombination on the neutral evolution of genetic ...
Efficient Natural Evolution Strategies - Idsia
Adaptive evolution of digestive RNASE1 genes in leaf-eating ...
Research article Adaptive evolution of the insulin gene in ...
Evolution Strategies for Constants Optimization in Genetic ...
Adaptive evolution of energy metabolism genes and the origin of ...
Extensive gene gain associated with adaptive evolution of poxviruses
Accelerated regulatory gene evolution in an adaptive radiation
Link Adaptation Strategy for IEEE 802.11 WLAN via ... - CiteSeer
SPECTRAL ANALYSIS USING EVOLUTION STRATEGIES
Adaptation and the evolution of parasite virulence in a connected ...
Recent acceleration of human adaptive evolution - University of Utah