The computational intelligence (CI) paradigm [1] is a triumvirate of three technologies—neural networks [2, 3], fuzzy logic [4], and evolutionary algorithms [5–7]—which stresses their seamless integration, resulting in numerous important spin-off commer

ParallelComputationalIntelligenceModels

ProfessorSatishKumar,SeniorMember,IEEE

DepartmentofPhysicsandComputerScience

Dr.SandeepPaul

DepartmentofElectricalEngg.(TechnicalCollege)

DayalbaghEducationalInstitute

Dayalbagh,Agra,INDIAskumardb@ieee.org,spaul.dei@gmail.com

SpecialSessionProposalfor

WorldCongressonComputationalIntelligence

HongKong,2008

1OutlineandMotivation

Thecomputationalintelligence(CI)paradigm[1]isatriumvirateofthreetechnologies—neuralnetworks[2,3],fuzzylogic[4],andevolutionaryalgorithms[5–7]—whichstressestheirseamlessintegration,resultinginnumerousimportantspin-o commercialapplications.Theintegra-tionofthesetechnologieshasassumedvariousformssuchasneuro-genetic,geneticfuzzyandevolutionary-neuro-fuzzyhybrids.SinceEA’so ere cientsolutionsforsystemoptimization,itisnotsurprisingthatthesealgorithmsarenowincreasinglyemployedforautomaticsimul-taneousestimationofparameters,architectureidenti cationofCIhybrids,andeliminationofredundantfeaturesthroughglobalsearch[8–10].Forexample,inthequestofobtaininganoptimalnetworkarchitectures,featureselection[9,11]andantecedentandconsequentconnec-tivity[12,13]thesehavealsobeenexploredtoagreatextent.Theextensiverequirementofcomputationalresourcesforstring tnessevaluationscoupledwiththelargenumberofsuchevaluationsrequiredduringiterativerunsofevolutionaryalgorithmsprovidesmotivationfortheirparallelimplementation,byexploitingtheintrinsicallyparallelnatureofEAs.References

[14,15]provideextensivereviewofparallelevolutionarymodels,parallelimplementations,andpressingtheoreticalissues.

Overthepastfewyears,clustercomputinghasenteredthemainstreamofhighperfor-mancecomputing.Owingtoitslowcost,highperformance,freelyavailablesoftwaretoolsfordevelopingclusterapplications,andtheadvantageofbeingabletouseexistingnetworksforcomputation,clustercomputinghasbecomeane ectivechoicetorunalltypesofparallelanddistributedapplications.VariouslibrariesandtoolsforparallelimplementationsincludetheParallelVirtualMachine(PVM)[16],MessagePassingInterface(MPI)[17,18],andGlobus

[19].Ofthese,MPIhasemergedasthedefactostandardforparallelimplementationstoday.SomeexamplesparallelimplementationsofCIhybridmodelsintheliteratureinclude:PARGE-FREX[20]whichisaparallelversionofGEFREX[21]andemploysPVMasitsimplementationtool;[22],whichisaparallelversionoflearninginasimilaritybasedevolutionary-neuro-fuzzy

1