Abstract. Aliasing environments present the tasks of increased difficulty for Learning Classifier Systems. Aliasing squares look identical for an agent with limited perceptive power, but may demand a completely different optimal strategy. Thus, the presenc
AgentPmodel:LearningClassi erSystemwith
associativeperception
ZhannaVolodymyrivnaZatuchna
TechnicalReportCMP-C04-03
SchoolofComputingSciences,
UniversityofEastAnglia,
Norwich,NR47TJ,England
z.zatuchna@uea.ac.uk
Abstract.Aliasingenvironmentspresentthetasksofincreaseddi -cultyforLearningClassi erSystems.Aliasingsquareslookidenticalforanagentwithlimitedperceptivepower,butmaydemandacompletelydi erentoptimalstrategy.Thus,thepresenceofaliasingsquaresinamazemayleadtoanon-optimalbehaviouranddecreasetheagent’sper-formance.Asapossibleapproachtotheproblemweintroduceapsycho-logicalmodelofassociativeperceptionlearningandbasedonthemodelAgentP,anLCSwithexplicitlyimprintedimagesoftheenvironmentalstates.Thesystemistestedonafewaliasingandnon-aliasingenviron-mentstoestablishthelearninge ectivenessoftheapproach.
Keywords:AgentP,LearningClassi erSystems,LCS,associativeperception,aliasing,maze,environment.
1Introduction
LearningClassi erSystems(LCS)belongtotheclassofsystemsbasedontheprincipleofself-organizationandevolutionadoptedfromrealnature.Thecom-monlyusedproblemtoestimateLCSperformanceisamazeenvironment.Theagentdoesnothaveamodelofthemazeandhastolearnhowtobehavebasedonalimitedfeedbackfromtheenvironment.AnLCSisanalgorithmforprocessingthefeedbackandselectinganactionbasedonasetofbehaviouralrules.
Themazeswithaliasingsquares,thatlookidenticalforanagentwithlimitedperceptivepower,butaredi erentinfact,presentataskofincreaseddi cultyforLCS.AsapossibleapproachtotheproblemweintroducetheAgentPsystem,aLearningClassi erSystemwithassociativeperception,thathasinheriteditsnamefromthepsychologicallearningmodel,alsointroducedinthereport.Thenextsectionprovidesasummaryofrelatedworkandisfollowedbythedescriptionoftheassociativeperceptionlearningmodel(Section4).ThenadescriptionofthestructureandperformanceoftheAgentPsystemfollows.Section5informsabouttheconductedexperimentsandtheobtainedresults.Next,theareasoffutureworkareoutlined.Finally,conclusionsareprovided.
AUTO-ASSOCIATIVE NEURAL NETWORK SYSTEM FOR ...classifier which the outputs depend on the inputs...Autoencoder networks with self-supervised learning ...
ABSTRACT An Associative ... 暂无评价 6页 免费... The perception of melodi... 18页 免费 The A...We then queried the classifier with the second ...
ABSTRACT An Associative ... 暂无评价 6页 免费... The perception of melodi... 18页 免费 The A...We then queried the classifier with the second ...
热门文档