Home

Program

Search

Author Index

Sponsors

Committee

Contact Us

CD Tech Support

 

 

Session:

Workshop - SOEA

Title:

On Location Independent Representations and Self-Organization

   

Authors:

Ivan Garibay
Annie S. Wu
Ozlem Garibay

   

Abstract:

We study the self-organization of genomic symbols on agenetic algorithm with a location independent representation·the Proportional Genetic Algorithm (PGA) [2]. Self-organizationof genomic symbols is possible because location independentrepresentations ensure the absence of selective pressurefor a particular order. We hypothesize that self-similarityemerges because self-similar genomes are more robust withrespect to crossover and mutation and because it favors positivecorrelations between the form and quality of candidatesolutions.The PGA is a Genetic Algorithm (GA) with a representationbased on protein concentrations rather than on theusual gene ordering. A PGA translates strings of genes intomultisets of proteins prior to fitness evaluation. As a result,there is no fitness pressure for any particular gene orderingand the order of the genes is free to evolve along withthe candidate solutions that they encode. Previous studieshave shown that genomic symbols under these circumstancesare evenly distributed throughout the genome andthat they appear to form building blocks of a peculiar type:coarse grained versions of the entire genome. We ask thefundamental questions: what is the emergent genomic orderingwhen there is no selective pressure for any particularordering? and why?. We use two very different methodsto analyse the emergent genomic structure: standard equalsymbolcorrelation analysis, and an experimental method ofour own making to analyse the self-similarity of genomic segmentswith repect to fitness. Our results can be summarizedas follows:1. The equal-symbol correlation on completely locationindependent genomes, as implemented by the PGA,resembles white noise behavior and the emergent genomicstructure is self-similar with respect to fitness.2. Emergent genomic self-similarity seems to produce thefollowing effect: it favors positive correlations betweenform and quality of candidate solutions, a key propertyneeded for stochastic search algorithms such as evolutionaryalgorithms; and it reduces schemata disruptioncaused by crossover.

CD-ROM Produced by X-CD Technologies