An Experimental Application of the Learnable Evolution Model and Genetic Algorithms to Parameter Estimation in Digital Signal Filters Design
Mark A. Coletti, Thomas D. Lash, Craig Mandsager, Moustafa, Rida E., Ryszard S. Michalski · 1999
ThispaperdescribesanapplicationofLEM1,apreliminaryimplementationof Learnable EvolutionModel(LEM),andtwocanonicalgeneti calgorithms,GA1andGA2,toparameter estimationin digitalsignalfilter design . LEM1alternatesbetweentwomodesofoperation: MachineLearningmode,whichemploysAQ-18rulelearningsystem,andDarwinianEvolution mode,whichemploysgeneticalgorit hmGA2.MachineLearningmodegenerateshypothesesas towhattypeofindividualsinapopulationrepresenthighfitnesssolutions.Thesehypotheses, expressedintheformofattributionalrules,areusedtogeneratenewpopulationsofsolutions. Whenthe topfitnessofapopulationhasnotimprovedsufficientlyduringonemode,LEM1 switchestoanothermode.LEM1alternatesbetweenthetwomodesuntilaglobaltermination conditionissatisfied.Intheexperiments ,LEM-1significantlyoutperformedgenetic algorithms GA1andGA2. Keywords:Evolutionarycomputation,LearnableEvolutionModel,GeneticAlgorithms, FunctionOptimization,SymbolicLearning,AQ18,DigitalFilters. 1