Instant ebooks textbook Cybernetical intelligence: engineering cybernetics with machine intelligence

Page 1


Machine Intelligence

Visit to download the full and correct content document: https://ebookmass.com/product/cybernetical-intelligence-engineering-cybernetics-with -machine-intelligence-kelvin-k-l-wong/

More products digital (pdf, epub, mobi) instant download maybe you interests ...

Human-Like Machine Intelligence Stephen Muggleton

https://ebookmass.com/product/human-like-machine-intelligencestephen-muggleton/

Visualization Techniques for Climate Change with Machine Learning and Artificial Intelligence Arun Lal Srivastav

https://ebookmass.com/product/visualization-techniques-forclimate-change-with-machine-learning-and-artificial-intelligencearun-lal-srivastav/

Engineering Intelligent Systems: Systems Engineering and Design with Artificial Intelligence, Visual Modeling Barclay R. Brown

https://ebookmass.com/product/engineering-intelligent-systemssystems-engineering-and-design-with-artificial-intelligencevisual-modeling-barclay-r-brown/

Machine Intelligence, Big Data Analytics, and IoT in Image Processing Ashok Kumar

https://ebookmass.com/product/machine-intelligence-big-dataanalytics-and-iot-in-image-processing-ashok-kumar/

Artificial Intelligence and Machine Learning for EDGE Computing 1st Edition Rajiv Pandey

https://ebookmass.com/product/artificial-intelligence-andmachine-learning-for-edge-computing-1st-edition-rajiv-pandey/

Applications of Artificial Intelligence in Process Systems Engineering Jingzheng Ren

https://ebookmass.com/product/applications-of-artificialintelligence-in-process-systems-engineering-jingzheng-ren/

Intelligence Science: Leading the Age of Intelligence Zhongzhi Shi

https://ebookmass.com/product/intelligence-science-leading-theage-of-intelligence-zhongzhi-shi/

Sustainable Developments by Artificial Intelligence and Machine Learning for Renewable Energies Krishna Kumar

https://ebookmass.com/product/sustainable-developments-byartificial-intelligence-and-machine-learning-for-renewableenergies-krishna-kumar/

Forecasting with Artificial Intelligence: Theory and Applications Mohsen Hamoudia

https://ebookmass.com/product/forecasting-with-artificialintelligence-theory-and-applications-mohsen-hamoudia/

CyberneticalIntelligence

IEEEPress

445HoesLane Piscataway,NJ08854

IEEEPressEditorialBoard

SarahSpurgeon, EditorinChief

JónAtliBenediktssonBehzadRazaviJeffreyReed AnjanBoseJimLykeDiomidisSpinellis JamesDuncan AminMoeness DesineniSubbaramNaidu

HaiLi BrianJohnson AdamDrobot TomRobertazzi AhmetMuratTekalp

CyberneticalIntelligence

EngineeringCyberneticswithMachineIntelligence

KelvinK.L.Wong

Copyright©2024byTheInstituteofElectricalandElectronicsEngineers,Inc.Allrightsreserved.

PublishedbyJohnWiley&Sons,Inc.,Hoboken,NewJersey. PublishedsimultaneouslyinCanada.

Nopartofthispublicationmaybereproduced,storedinaretrievalsystem,ortransmittedinany formorbyanymeans,electronic,mechanical,photocopying,recording,scanning,orotherwise, exceptaspermittedunderSection107or108ofthe1976UnitedStatesCopyrightAct,without eitherthepriorwrittenpermissionofthePublisher,orauthorizationthroughpaymentofthe appropriateper-copyfeetotheCopyrightClearanceCenter,Inc.,222RosewoodDrive,Danvers, MA01923,(978)750-8400,fax(978)750-4470,oronthewebatwww.copyright.com.Requeststo thePublisherforpermissionshouldbeaddressedtothePermissionsDepartment,JohnWiley& Sons,Inc.,111RiverStreet,Hoboken,NJ07030,(201)748-6011,fax(201)748-6008,oronlineat http://www.wiley.com/go/permission.

Trademarks:WileyandtheWileylogoaretrademarksorregisteredtrademarksofJohnWiley& Sons,Inc.and/oritsaffiliatesintheUnitedStatesandothercountriesandmaynotbeusedwithout writtenpermission.Allothertrademarksarethepropertyoftheirrespectiveowners.JohnWiley& Sons,Inc.isnotassociatedwithanyproductorvendormentionedinthisbook.

LimitofLiability/DisclaimerofWarranty:Whilethepublisherandauthorhaveusedtheirbest effortsinpreparingthisbook,theymakenorepresentationsorwarrantieswithrespecttothe accuracyorcompletenessofthecontentsofthisbookandspecificallydisclaimanyimplied warrantiesofmerchantabilityorfitnessforaparticularpurpose.Nowarrantymaybecreatedor extendedbysalesrepresentativesorwrittensalesmaterials.Theadviceandstrategiescontained hereinmaynotbesuitableforyoursituation.Youshouldconsultwithaprofessionalwhere appropriate.Neitherthepublishernorauthorshallbeliableforanylossofprofitoranyother commercialdamages,includingbutnotlimitedtospecial,incidental,consequential,orother damages.Further,readersshouldbeawarethatwebsiteslistedinthisworkmayhavechangedor disappearedbetweenwhenthisworkwaswrittenandwhenitisread.Neitherthepublishernor authorsshallbeliableforanylossofprofitoranyothercommercialdamages,includingbutnot limitedtospecial,incidental,consequential,orotherdamages.

Forgeneralinformationonourotherproductsandservicesorfortechnicalsupport,pleasecontact ourCustomerCareDepartmentwithintheUnitedStatesat(800)762-2974,outsidetheUnited Statesat(317)572-3993orfax(317)572-4002.

Wileyalsopublishesitsbooksinavarietyofelectronicformats.Somecontentthatappearsinprint maynotbeavailableinelectronicformats.FormoreinformationaboutWileyproducts,visitour websiteatwww.wiley.com.

LibraryofCongressCataloging-in-PublicationDataappliedfor:

Hardback:9781394217489

CoverDesign:Wiley

CoverImage:Courtesyofauthor

Setin9.5/12.5ptSTIXTwoTextbyStraive,Pondicherry,India

Contents

Preface xv

AbouttheAuthor xix

AbouttheCompanionWebsite xxi

1ArtificialIntelligenceandCyberneticalLearning 1

1.1ArtificialIntelligenceInitiative 1

1.2IntelligentAutomationInitiative 4

1.2.1BenefitsofIAI 5

1.3ArtificialIntelligenceVersusIntelligentAutomation 5

1.3.1ProcessDiscovery 6

1.3.2Optimization 7

1.3.3AnalyticsandInsight 8

1.4TheFourthIndustrialRevolutionandArtificialIntelligence 9

1.4.1ArtificialNarrowIntelligence 10

1.4.2ArtificialGeneralIntelligence 12

1.4.3ArtificialSuperIntelligence 13

1.5PatternAnalysisandCognitiveLearning 14

1.5.1MachineLearning 15

1.5.1.1ParametricAlgorithms 16

1.5.1.2NonparametricAlgorithms 17

1.5.2DeepLearning 20

1.5.2.1ConvolutionalNeuralNetworksinAdvancingArtificialIntelligence 21

1.5.2.2FutureAdvancementinDeepLearning 22

1.5.3CyberneticalLearning 23

1.6CyberneticalArtificialIntelligence 24

1.6.1ArtificialIntelligenceControlTheory 24

1.6.2InformationTheory 26

1.6.3CyberneticSystems 27

1.7CyberneticalIntelligenceDefinition 28

1.8TheFutureofCyberneticalIntelligence 30

Summary 32

ExerciseQuestions 32 FurtherReading 33

2CyberneticalIntelligentControl 35

2.1ControlTheoryandFeedbackControlSystems 35

2.2Maxwell’sAnalysisofGovernors 37

2.3HaroldBlack 39

2.4NyquistandBode 40

2.5StaffordBeer 42

2.5.1CyberneticControl 42

2.5.2ViableSystemsModel 42

2.5.3CyberneticsModelsofManagement 43

2.6JamesLovelock 43

2.6.1CyberneticApproachtoEcosystems 43

2.6.2GaiaHypothesis 44

2.7MacyConference 44

2.8McCulloch –Pitts 45

2.9JohnvonNeumann 47

2.9.1DiscussionsonSelf-ReplicatingMachines 47

2.9.2DiscussionsonMachineLearning 48

Summary 48

ExerciseQuestions 49 FurtherReading 50

3TheBasicsofPerceptron 51

3.1TheAnalogyofBiologicalandArtificialNeurons 51

3.1.1BiologicalNeuronsandNeurodynamics 52

3.1.2TheStructureofNeuralNetwork 53

3.1.3EncodingandDecoding 56

3.2PerceptionandMultilayerPerceptron 57

3.2.1BackPropagationNeuralNetwork 59

3.2.2DerivativeEquationsforBackpropagation 59

3.3ActivationFunction 61

3.3.1SigmoidActivationFunction 61

3.3.2HyperbolicTangentActivationFunction 62

3.3.3RectifiedLinearUnitActivationFunction 62

3.3.4LinearActivationFunction 64

Summary 65

ExerciseQuestions 67 FurtherReading 67

4TheStructureofNeuralNetwork 69

4.1LayersinNeuralNetwork 69

4.1.1InputLayer 69

4.1.2HiddenLayer 70

4.1.3Neurons 70

4.1.4WeightsandBiases 71

4.1.5ForwardPropagation 72

4.1.6Backpropagation 72

4.2PerceptronandMultilayerPerceptron 73

4.3RecurrentNeuralNetwork 75

4.3.1LongShort-TermMemory 76

4.4MarkovNeuralNetworks 77

4.4.1StateTransitionFunction 77

4.4.2ObservationFunction 78

4.4.3PolicyFunction 78

4.4.4LossFunction 78

4.5GenerativeAdversarialNetwork 78 Summary 79 ExerciseQuestions 80 FurtherReading 81

5BackpropagationNeuralNetwork 83

5.1BackpropagationNeuralNetwork 83

5.1.1ForwardPropagation 85

5.2GradientDescent 85

5.2.1LossFunction 85

5.2.2ParametersinGradientDescent 88

5.2.3GradientinGradientDescent 88

5.2.4LearningRateinGradientDescent 89

5.2.5UpdateRuleinGradientDescent 89

5.3StoppingCriteria 89

5.3.1ConvergenceandStoppingCriteria 90

5.3.2LocalMinimumandGlobalMinimum 91

5.4ResamplingMethods 91

5.4.1Cross-Validation 93

5.4.2Bootstrapping 93

5.4.3MonteCarloCross-Validation 94

5.5OptimizersinNeuralNetwork 94

5.5.1StochasticGradientDescent 94

5.5.2RootMeanSquarePropagation 96

5.5.3AdaptiveMomentEstimation 96

5.5.4AdaMax 97

5.5.5MomentumOptimization 97

Summary 97

ExerciseQuestions 99 FurtherReading 100

6ApplicationofNeuralNetworkinLearningandRecognition 101

6.1ApplyingBackpropagationtoShapeRecognition 101

6.2SoftmaxRegression 105

6.3 K-BinaryClassifier 107

6.4RelationalLearningviaNeuralNetwork 108

6.4.1GraphNeuralNetwork 109

6.4.2GraphConvolutionalNetwork 111

6.5CyberneticsUsingNeuralNetwork 112

6.6StructureofNeuralNetworkforImageProcessing 115

6.7TransformerNetworks 116

6.8AttentionMechanisms 116

6.9GraphNeuralNetworks 117

6.10TransferLearning 118

6.11GeneralizationofNeuralNetworks 119

6.12PerformanceMeasures 120

6.12.1ConfusionMatrix 120

6.12.2ReceiverOperatingCharacteristic 121

6.12.3AreaUndertheROCCurve 122

Summary 123

ExerciseQuestions 123 FurtherReading 124

7CompetitiveLearningandSelf-OrganizingMap 125

7.1PrincipalofCompetitiveLearning 125

7.1.1Step1:NormalizedInputVector 128

7.1.2Step2:FindtheWinningNeuron 128

7.1.3Step3:AdjusttheNetworkWeightVectorandOutputResults 129

7.2BasicStructureofSelf-OrganizingMap 129

7.2.1PropertiesSelf-OrganizingMap 130

7.3Self-OrganizingMappingNeuralNetworkAlgorithm 131

7.3.1Step1:InitializeParameter 132

7.3.2Step2:SelectInputsandDetermineWinningNodes 132

7.3.3Step3:AffectNeighboringNeurons 132

7.3.4Step4:AdjustWeights 133

7.3.5Step5:JudgingtheEndCondition 133

7.4GrowingSelf-OrganizingMap 133

7.5TimeAdaptiveSelf-OrganizingMap 136

7.5.1TASOM-BasedAlgorithmsforRealApplications 138

7.6OrientedandScalableMap 139

7.7GenerativeTopographicMap 141

Summary 145 ExerciseQuestions 146 FurtherReading 147

8SupportVectorMachine 149

8.1TheDefinitionofDataClustering 149

8.2SupportVectorandMargin 152

8.3KernelFunction 155

8.3.1LinearKernel 155

8.3.2PolynomialKernel 156

8.3.3RadialBasisFunction 157

8.3.4LaplaceKernel 159

8.3.5SigmoidKernel 159

8.4LinearandNonlinearSupportVectorMachine 160

8.5HardMarginandSoftMargininSupportVectorMachine 164

8.6I/OofSupportVectorMachine 167

8.6.1TrainingData 167

8.6.2FeatureMatrixandLabelVector 168

8.7HyperparametersofSupportVectorMachine 169

8.7.1TheCHyperparameter 169

8.7.2KernelCoefficient 169

8.7.3ClassWeights 170

8.7.4ConvergenceCriteria 170

8.7.5Regularization 171

8.8ApplicationofSupportVectorMachine 171

8.8.1Classification 171

8.8.2Regression 173

8.8.3ImageClassification 173

8.8.4TextClassification 174

Summary 174 ExerciseQuestions 175 FurtherReading 176

9Bio-InspiredCyberneticalIntelligence 177

9.1GeneticAlgorithm 178

9.2AntColonyOptimization 181

x Contents

9.3BeesAlgorithm 184

9.4ArtificialBeeColonyAlgorithm 186

9.5CuckooSearch 189

9.6ParticleSwarmOptimization 193

9.7BacterialForagingOptimization 196

9.8GrayWolfOptimizer 197

9.9FireflyAlgorithm 199

Summary 200

ExerciseQuestions 201

FurtherReading 202

10Life-InspiredMachineIntelligenceandCybernetics 203

10.1Multi-AgentAISystems 203

10.1.1GameTheory 205

10.1.2DistributedMulti-AgentSystems 206

10.1.3Multi-AgentReinforcementLearning 207

10.1.4EvolutionaryComputationandMulti-AgentSystems 209

10.2CellularAutomata 211

10.3DiscreteElementMethod 212

10.3.1Particle-BasedSimulationofBiologicalCellsandTissues 214

10.3.2SimulationofMicrobialCommunitiesandTheirInteractions 215

10.3.3DiscreteElementMethod-BasedModelingofBiologicalFluidsand SoftMaterials 216

10.4SmoothedParticleHydrodynamics 218

10.4.1SPH-BasedSimulationsofBiomimeticFluidDynamic 219

10.4.2SPH-BasedSimulationsofBio-InspiredEngineeringApplications 220

Summary 221

ExerciseQuestions 222 FurtherReading 223

11RevisitingCyberneticsandRelationtoCyberneticalIntelligence 225

11.1TheConceptandDevelopmentofCybernetics 225

11.1.1AttributesofControlConcepts 225

11.1.2ResearchObjectsandCharacteristicsofCybernetics 226

11.1.3DevelopmentofCyberneticalIntelligence 227

11.2TheFundamentalIdeasofCybernetics 227

11.2.1SystemIdea 227

11.2.2InformationIdea 229

11.2.3BehavioralIdea 230

11.2.4CyberneticalIntelligenceNeuralNetwork 231

11.3CyberneticExpansionintoOtherFieldsofResearch 234

11.3.1SocialCybernetics 234

11.3.2InternalControl-RelatedTheories 237

11.3.3SoftwareControlTheory 237

11.3.4PerceptualCybernetics 238

11.4PracticalApplicationofCybernetics 240

11.4.1ResearchontheControlMechanismofNeuralNetworks 240

11.4.2BalanceBetweenInternalControlandManagementPower Relations 240

11.4.3SoftwareMarkovAdaptiveTestingStrategy 242

11.4.4TaskAnalysisModel 244

Summary 245

ExerciseQuestions 246

FurtherReading 247

12TuringMachine 249

12.1BehaviorofaTuringMachine 250

12.1.1ComputingwithTuringMachines 251

12.2BasicOperationsofaTuringMachine 252

12.2.1ReadingandWritingtotheTape 253

12.2.2MovingtheTapeHead 254

12.2.3ChangingStates 254

12.3InterchangeabilityofProgramandBehavior 255

12.4ComputabilityTheory 256

12.4.1ComplexityTheory 257

12.5AutomataTheory 258

12.6PhilosophicalIssuesRelatedtoTuringMachines 259

12.7HumanandMachineComputations 260

12.8HistoricalModelsofComputability 261

12.9RecursiveFunctions 262

12.10TuringMachineandIntelligentControl 263 Summary 264

ExerciseQuestions 265 FurtherReading 265

13EntropyConceptsinMachineIntelligence 267

13.1RelativeEntropyofDistributions 268

13.2RelativeEntropyandMutualInformation 268

13.3EntropyinPerformanceEvaluation 269

13.4Cross-EntropySoftmax 271

13.5CalculatingCross-Entropy 272

13.6Cross-EntropyasaLossFunction 273

13.7Cross-EntropyandLogLoss 274

13.8ApplicationofEntropyinIntelligentControl 275

13.8.1Entropy-BasedControl 275

13.8.2FuzzyEntropy 276

13.8.3Entropy-BasedControlStrategies 277

13.8.4Entropy-BasedDecision-Making 278

Summary 279

ExerciseQuestions 279

FurtherReading 280

14SamplingMethodsinCyberneticalIntelligence 283

14.1IntroductiontoSamplingMethods 283

14.2BasicSamplingAlgorithms 284

14.2.1ImportanceofSamplingMethodsinMachineIntelligence 286

14.3MachineLearningSamplingMethods 287

14.3.1RandomOversampling 288

14.3.2RandomUndersampling 290

14.3.3SyntheticMinorityOversamplingTechnique 290

14.3.4AdaptiveSyntheticSampling 292

14.4AdvantagesandDisadvantagesofMachineLearningSampling Methods 293

14.5AdvancedSamplingMethodsinCyberneticalIntelligence 294

14.5.1EnsembleSamplingMethod 295

14.5.2ActiveLearning 297

14.5.3BayesianOptimizationinSampling 299

14.6ApplicationsofSamplingMethodsinCyberneticalIntelligence 302

14.6.1ImageProcessingandComputerVision 302

14.6.2NaturalLanguageProcessing 304

14.6.3RoboticsandAutonomousSystems 307

14.7ChallengesandFutureDirections 308

14.8ChallengesandLimitationsofSamplingMethods 309

14.9EmergingTrendsandInnovationsinSamplingMethods 309

Summary 310

ExerciseQuestions 311

FurtherReading 312

15DynamicSystemControl 313

15.1LinearSystems 314

15.2NonlinearSystem 316

15.3StabilityTheory 318

15.4ObservabilityandIdentification 320

15.5ControllabilityandStabilizability 321

15.6OptimalControl 323

15.7LinearQuadraticRegulatorTheory 324

15.8Time-OptimalControl 326

15.9StochasticSystemswithApplications 328

15.9.1StochasticSysteminControlSystems 329

15.9.2StochasticSysteminRoboticsandAutomation 329

15.9.3StochasticSysteminNeuralNetworks 330 Summary 331

ExerciseQuestions 331 FurtherReading 332

16DeepLearning 333

16.1NeuralNetworkModelsinDeepLearning 335

16.2MethodsofDeepLearning 336

16.2.1ConvolutionalNeuralNetworks 337

16.2.2RecurrentNeuralNetworks 340

16.2.3GenerativeAdversarialNetworks 342

16.2.4DeepLearningBasedImageSegmentationModels 345

16.2.5VariationalAutoEncoders 348

16.2.6TransformerModels 350

16.2.7Attention-BasedModels 352

16.2.8Meta-LearningModels 354

16.2.9CapsuleNetworks 357

16.3DeepLearningFrameworks 358

16.4ApplicationsofDeepLearning 359

16.4.1ObjectDetection 360

16.4.2IntelligentPowerSystems 361

16.4.3IntelligentControl 362 Summary 362

ExerciseQuestions 363 References 364 FurtherReading 365

17NeuralArchitectureSearch 367

17.1NeuralArchitectureSearchandNeuralNetwork 369

17.2ReinforcementLearning-BasedNeuralArchitectureSearch 371

17.3EvolutionaryAlgorithms-BasedNeuralArchitectureSearch 374

17.4BayesianOptimization-BasedNeuralArchitectureSearch 376

17.5Gradient-BasedNeuralArchitectureSearch 378

17.6One-shotNeuralArchitectureSearch 379

17.7Meta-Learning-BasedNeuralArchitectureSearch 381

17.8NeuralArchitectureSearchforSpecificDomains 383

17.8.1CyberneticalIntelligentSystems:NeuralArchitectureSearchin Real-World 384

17.8.2NeuralArchitectureSearchforSpecificCyberneticalControlTasks 385

17.8.3NeuralArchitectureSearchforCyberneticalIntelligentSystemsin Real-World 386

17.8.4NeuralArchitectureSearchforAdaptiveCyberneticalIntelligent Systems 388

17.9ComparisonofDifferentNeuralArchitectureSearchApproaches 389 Summary 391

ExerciseQuestions 391

FurtherReading 392

FinalNoteson CyberneticalIntelligence 393 Index 399

Preface

Lifeevolvesintoexistencefromtheedgeofchaosandbuildsupanewmechanism basedonasetofrulesthatgovernsthelawofsurvival,reproduction,andevolution.Thiscomplexsetofrulesthatallowsalivingthingtointeractwithitsenvironmentcomesintobeingfromthebeginningoflifeitself,whichissomethingone canunderstandasintelligence.Allthewondersofart,design,sciences,etc.,inthe worldmakeusponderuponthequestionoftheages,ontheoriginofcreationand theexistenceoflifeitselfandtheevolutionofintelligencethatcomesintobeing.

“Theimportantthingisnottostopquestioning,” isafamousquotationbyAlbert Einsteinfrom1955.Arobotgainsintelligencebyquestioningandseekinganswers, whichformexamplesforthelabelsfortheunseenexamples.However,willthe robothaveacreativemindsimilartothatofEinstein?Thisleadsustothequestion ofwhethercreativitycanbeprogrammed.Canananalogybebridgedbetweenthe robot’sexperienceindevelopingmultiplesearchwaypathsfortheoptimalsolutionandanintelligentbeing’sintuitiontodesigncreativesolutions?

Artificialintelligence(AI)isbuiltonthepillarsofafewmajorbranchesofscienceandengineering,namely,systematology,informationtheory,andcybernetics,whichistypicallybasedoncontroltheorythatwasderivedfromthestudiesof NorbertWiener,theworld-renownedfatherofcybernetics.In1954,Hsue-sen Tsienfoundedengineeringcyberneticsbypublishingthefamousengineering cyberneticsinAmerica.Onthebasisofcybernetics,apredictivesystemmaybe regardedasamultiplefeedbacksystem.Theframeworkofamultilayerperceptron aswellasthatofabackpropagationneuralnetworkcanbebasedonthetheoretic ofsystemcontrolinmoderncybernetics.Withthistypeofthinking,perceptron theoryoffersacohesiveapproachtothestatisticalmechanicsandprinciplesof cyberneticsasabasisforthesuccessfulneuralnetworkmodeling.

Afeedbackcontroller’soperationistochangethebehaviorofasystemfundamentally.Feedbackcontrolsystemssampleasystem’soutputs,comparethemtoa setofdesiredoutputs,andthenutilizetheresultingerrorsignalstocomputethe

system’scontrolinputsinsuchawaythattheerrorsareminimized.Artificially builtfeedbackcontrolsystems,whichareutilizedtogovernindustrial,automotive,andaeronauticalsystems,areresponsiblefortoday’saerospaceachievements. Biologicalsystemsarefullofnaturallyoccurringfeedbackcontrols.Thecell,oneof themostbasicofalllifeforms,regulatesthepotentialdifferenceacrossthecell membranetopreservehomeostasis.Althoughneuralnetworkcontrollersare adaptivelearningsystems,theydonotneedtheconventionalassumptionsof adaptivecontroltheory,suchasparameterlinearityandthepresencearegression matrix.Itisdemonstratedindetailtheprocesstocreateneuralnetworkbasedcontrollersforcyberneticalsystems,ageneralcategoryofnonlinearsystems,complicatedindustrialsystemswithvibrationsandflexibilityeffects,forcecontrol,motor dynamicscontrol,andotherapplications.Thesearegivenforbothcontinuoustimeanddiscrete-timeweighttuning.

IntegrationofAIandcyberneticscanproduceapplicationsinpredictivecontrol, patternrecognition,andclassification,whichessentiallyarebasedonthesame fundamentals.Thisbookproposesforthefirsttimethenovelperspectiveof machineintelligence,whichistermedas CyberneticalIntelligence.Suchanewfield willhaveextensiveandpracticalapplicationsinnotjustthecombinatorialoptimizationproblemsbutalsoinpatternrecognition,datamining,andotherrelated machineintelligencebasedcyberneticsproblems.

Thekeyconceptof CyberneticalIntelligence grewfromadesiretounderstand andbuildsystemsthatcanachievegoals,whethercomplexhumangoalsorjust goals.Itisevendeeperunderlyingconceptualterm.Cyberneticsholdstheworld sufficientlytogainfeedbackinordertocorrecttheactionstoachievegoals.Itis mutualcombinationofautomatedcommunicationandcontrolsystembetween artificiallyintelligentmachinesandtheenvironmentwithsubsequentstrongsupportfrommachinelearning;theconceptsofsystemsthinkingandsystemstheory becameintegralpartsoftheestablishedscientificlanguageof CyberneticalIntelligence andcanleadtonumerousnewmethodologiesandapplications.

Thebasicideasof CyberneticalIntelligence canbetreatedwithoutreferenceto electronics,buttheyarefundamentallychallenging;soalthoughadvancedtechniquesmaybenecessaryforadvancedapplications,agreatdealcanbedone,especiallyinbiologicalsciences,bytheuseofmathematicalderivations,providedthey areusedwithaclearanddeepunderstandingoftheprincipalsinvolved.

Thisbookisintendedtoprovideaconciseconceptualizationof Cybernetical Intelligence .Itstartsfromcommonplaceandwell-understoodconceptsandproceeds,stepbystep,toshowhowtheseconceptscanbemadeexactandhowthey canbedevelopeduntiltheyleadintosubjectssuchasfeedback,stability,regulation,ultrastability,information,coding,noise,andothercybernetictopics.Closedloopapplicationsandfeaturesofneuralnetworkareexaminedanddevelopedin greatdetailinthisbook,employingmathematicalstabilityproofapproachesthat

illustratehowtoconstructneuro-controllerswhilealsoensuringtheirstabilityand performance.Controlengineeringbasedconcepts,afamilyofmulti-loopneurocontrollersforvariousapplicationshavebeencreatedmethodically.

Therearestrategiesforbothcontinuous-timeanddiscrete-timeweighttuning given.Thebookisintendedforstudentstakingasecondsemesterincontroltheory,aswellasengineersinacademiaandindustrywhoconstructfeedbackcontrollersforcomplexsystemsfoundincommercial,industrial,andmilitary applications.Themanytypesofneuro-controllersareorganizedintablesforsimplereferencewhenitcomestodesignprocedures.

Thismaterialisacomprehensiveexplorationoftheadvancedterminologiesin AIandcybernetics.InChapter1,theconceptofAIanditsrelationtocybernetics areintroduced.Chapter2delvesintothetheoryofcyberneticalintelligenceand control.Chapter3coversthebasicsofperceptron,includingitsactivationfunction.Thestructureofthemultilayerperceptronneuralnetworkisdiscussedin Chapter4,whileChapter5coversthebackpropagationalgorithmanditsderivatives,aswellastheresamplingrate.Chapter6focusesonneuralnetworkapplicationsinlearningandrecognition.Chapter7exploresself-organizingandits applicationsinAI,andChapter8coverssupportvectormachinesandtheirapplications.Chapters9and10delveintobio-andlife-inspired CyberneticalIntelligence.Chapters11and12revisitcyberneticsanditsrelationto Cybernetical Intelligence andTuringmachines.Entropyconceptsandsamplingmethodsin CyberneticalIntelligence arecoveredinChapters13and14.Chapters15and16 describelinearsystemsanddeeplearning,includingtheirmethodsandapplications.Finally,Chapter17focusesonneuralarchitecturesearch,includingits methodsandapplications.Everychapterpresentsitsowncharacteristicconcept, andtheconcatenationoftheseconceptsgeneratesamindmapandgeneralframeworkfortheformulationofmachinelearningfromthecyberneticsperspectiveand encompassingthe CyberneticalIntelligence philosophy.Thephilosophicalinsights andmathematicaltheoriesinthisbookwillgiveustheadequateknowledgenecessaryforbuildingAI.

Itistheauthor’sbeliefthatthesubjectfoundediswellunderstoodandisthen builtupcarefully,stepbystep,withadvancedmathematical,computing,andengineeringknowledge.Havingspentyearsconsolidatinganddevelopingtheconceptualroadmapofmachinelearningfromthecyberneticsperspective,theauthoris proudtopresentthenovelworkon CyberneticalIntelligence totheacademiccommunitywiththeultimateaimoftrainingthenextgenerationofAIcybernetists.

AbouttheAuthor

Prof.Dr.KelvinK.L.Wongisadistinguishedexpert inmedicalimageprocessingandcomputational science,whoearnedhisPh.D.fromTheUniversity ofAdelaide.Withastrongacademicbackground fromNanyangTechnologicalUniversityandThe UniversityofSydney,hehasbeenattheforefront ofmergingthefieldsofcyberneticsandartificial intelligence(AI).Heiswidelyrecognizedforintroducingtheterm “CyberneticalIntelligence” and istheinventorandfounderoftheDeepRedAI system.Dr.Wong’simpactfulresearchinAIhas yieldedsignificantachievementswiththepotential topositivelyimpacthumanity.Heistheauthorofinfluentialbookssuchas MethodsinResearchandDevelopmentofBiomedicalDevices and Computational Hemodynamics–Theory,Modelling,andApplications.Withextensiveexperience asanassociateeditorandguesteditorforesteemedbiomedicalengineeringand computationalintelligencejournals,hehascontributedextensivelytothefield. AsaninternationallyrecognizedbiomedicalengineeringscientistandAI cybernetist,Dr.WongwasnamedamongStanfordUniversity’stop1.3%biomedicalengineeringresearchersin2020.Hehasactivelyparticipatedinresearching themanagementandcontrolofCOVID-19andisadedicatedsupporterand donortoUNICEF,advocatingforkindnessandhumanrights.Throughouthis professorship,hehasmentorednumerousstudents,providinginvaluableguidanceandshapingtheircareers.LeadingateamofexpertsinAI,healthcare, diseasemanagement,anddiagnosis,Dr.Wong’sexpertisehasbeeninstrumental insupportinggovernmentprojectsandinitiativesindevelopingcountries.

AbouttheCompanionWebsite

Thisbookisaccompaniedbyacompanionwebsite:

www.wiley.com/go/cyberintel

Thiswebsiteincludes: • AssignmentsandSolutions

ArtificialIntelligenceandCyberneticalLearning

Artificialintelligence(AI)isafieldofengineeringcyberneticsthatfocusesonthe developmentofintelligentmachinesthatcansimulatehuman-likebehaviorssuch aslearning,problem-solving,reasoning,anddecision-making.AItechnology involvestheuseofalgorithmsandcomputationalmodelstoanalyzevastamounts ofdata,recognizepatternsandmakepredictions,andinteractwithhumans throughnaturallanguageprocessing(NLP)andotherformsofcommunication. ThischapterwillcomprehensivelyexplorevariousaspectsofAI,includingits relationtocyberneticsandthefundamentalprinciplesgoverningit.Additionally, itwilldelveintothenuancesofparametricandnonparametricalgorithmsand coreconceptsofcyberneticalintelligence(CI).Throughasystematicandrigorous exposition,readerswillacquirearobustunderstandingofthekeyprinciplesand algorithmsthatunderlieAI.Consequently,theywillbewellequippedwiththe requisiteknowledgetodeveloptheirownAIapplications,leveragingtheinsights gainedfromthischapter.

1.1ArtificialIntelligenceInitiative

Intelligenceincludesthecapacityforabstraction,logic,learning,reasoning,communication,andinference.Itcanlearnfromtheenvironmentbothactivelyand passivelyandusetheknowledgetoobtainadaptiveability.AIcanbedefinedas ahuman-mademachinewithhuman-likeintelligence.TheuseofAIineducation hasproducedeffectivepedagogicaleffectsinadditiontotechnicaladvancements andtheoreticaldevelopments.Automatedtargetidentification,automaticmedical diagnosis,andaudiorecordingareafewinterestinguses.AImaybeutilizedto

CyberneticalIntelligence:EngineeringCyberneticswithMachineIntelligence,FirstEdition. KelvinK.L.Wong.

©2024TheInstituteofElectricalandElectronicsEngineers,Inc. Published2024byJohnWiley&Sons,Inc.

Companionwebsite:www.wiley.com/go/cyberintel

2 1ArtificialIntelligenceandCyberneticalLearning

providecustomizedassistanceandincreaseknowledge-gapawareness,allowing educatorstodeliverindividualizedandadaptableeducationwithefficiencyand effectiveness.Enablingcomputerstosimulateintelligentbehaviorusingprestored worldmodelsisthemaingoalofAI.TheAIsimulateshumancognitiveprocesses suchasreasoning,learning,patternrecognition,knowledgereasoning,and machinelearning(ML).MLreferstothecreationofautomatedsystemscapable ofprocessingmassivevolumesofdatafordataminingandisoneofthemoretraditionalfieldsofcomputingintelligence.

MLisapartofAIthatallowsmachinestoobtainintelligencefromdatawithout beingexplicitlyprogrammed.Therefore,itoftenassociatesMLwithdatamining. MLenablesacybersystemtopossessintelligencebyusingmassivedata.Basedon thatdata,MLmodelsoralgorithmscanminetheknowledge,rules,andlaws behindthedata.MLidentifiesunderlyingfunctionallinksinsystemsbetweensets ofvariablesandindividualvariables.ThegoalofcombiningthefieldsofMLand cyberneticsistoidentifydifferentwaysthatsystemsinteractwithoneanother throughvariousmethodsforlearningfromdata.Equation(1.1)illustrateshow MLmaybesummedupaslearningafunction(f)thatmapsinputvariables(x) tooutputvariables(y).

Theconfigurationofthefunctionisunknown,buttheMLalgorithmslearnto mapthetargetfunctionfromthetrainingdata.Itisnecessarytoassessmanyalgorithmstodeterminewhichoneisbestatmodelingtheunderlyingfunction becausetheyallreachdifferentconclusionsorexhibitbiasesonthefunction’s structure.Intheory,therearetwotypesofMLalgorithms:parametricalgorithms andnonparametricalgorithms.Additionally,threewell-knowntechniquesare usedtotrainMLalgorithms.Supervisedlearning,unsupervisedlearning,andreinforcementlearningarethethreecategoriesofML.

ThemostsignificantmethodologyinMLissupervisedlearning,whichis especiallycrucialintheprocessingofmultimediadata.Thiskindoflearning iscomparabletohowhumanslearnfromtheirpastexperiencestoobtainnew informationandimprovetheircapacitytocarryoutactivitiesintheactualworld. Modelsforsupervisedlearningaredesignedtopredicttheappropriatelabelfor newlypresenteddata.Unsupervisedlearningistypicallyusedtoidentifypatternsintheinputdatathatproposecandidatefeaturespriortotheapplication ofsupervisedlearning,andfeatureengineeringchangesthesecandidate featurestomakethemmoreappropriateforsupervisedlearning.Itisquite time-consumingtoidentifythecorrect categoryorresponseforeveryobservationinthetrainingsetinadditiontothecharacteristics.Withthehelpof semi-supervisedlearning,onemaytrainmodelswithverylittlelabeleddata, whichwillreducethelabelingwork.

1.1ArtificialIntelligenceInitiative 3

Unsupervisedlearningcanbemotiv atedfrominformation-theoreticand Bayesianprinciples.Itempowersthemodeltoworkindependentlytoidentify previouslyunnoticedpatternsandinformation.Takeintoaccountadevice(or livingthing)thatgetsaseriesofinputs,suchas x 1, x 2, …, x t ,where x t isthesensory inputattime t .Thisinput,whichisknownassensorydata,canbearepresentationofaretinalimage,acamera’ spixels,orasoundwaveform.Themost famoustechniqueisclusteringinwhicheachobservationbelongstoatleast oneofthe k clusters,while i and j belongtocentroidofeachcluster.Furthermore,variationwithineachclusterisachievedbyminimizingthesumofthe squaredEuclideandistancebetweeneachobservationwithinacluster,asshown inEquation(1.2).

where μk representsthecentroidofthe kth cluster, Xi isthe ith datapointinthe kth cluster,and Ck representsthesetofindicesofdatapointsassignedtothe kth cluster.Reinforcementlearning,ontheotherhand,isheavilyinfluencedbythetheory ofMarkovdecisionprocessesanddealswiththeabilitytolearntheassociations betweenstimuli,actions,andtheoccurrenceofpositiveevents.Theagentsare taughtarewardandpunishmentschemeinreinforcementlearning.Forwise actions,theagentisrewarded,andforpoorones,theyarepenalized.Whiledoing this,theagenttriestominimizetheundesirablemotionswhilemaximizingthe desirableones.Itishardlyunexpectedthatreinforcementlearninghasbeen noticedinthereallydistantpastgivenitsclearadaptivebenefit.Afewcybernetics experimentshavemadeuseofreinforcementlearning.Robotscanlearnskillsthat ahumaninstructorisunabletoteach,adaptalearnedabilitytoanewtask,and accomplishoptimizationevenintheabsenceofananalyticalformulationwith thehelpofthissortofML.Thepredictedtotaloftheimmediaterewardand thelong-termrewardunderthebestfeasiblepolicy(MaxPolicies),asgivenin Equation(1.3),isutility u (overalimitedagentlifespan):

where st isthestateattimestep t , R ( st , a )istheimmediaterewardofexecuting anactioninstate s t , N isthenumberofstepsinthelifetimeoftheagent,and R istherewardtimestep t .Theoperator E standsfortakinganexpectationover allsourcesofrandomnessinthesystem.Here, st denotesthestateattimestep t, R(st, a)istheinstantaneousbenefitofcarryingoutanactioninstate st,and N denotesthetotalnumberofstepstheagentwilltakethroughoutitslifespan. Takinganexpectationacrossallsystemrandomnesssourcesiswhattheoperator.

Theconfigurationofthefunctionisunknown,buttheMLalgorithmslearntomap thetargetfunctionfromthetrainingdata.Itisnecessarytocomparemultiplealgorithmstodeterminewhichoneisthebestsuccessfulatmodelingtheunderlying functionsincedifferentalgorithmsreachdifferentconclusionsorhavedifferent biasesonthestructureofthefunction.Asaresult,MLalgorithmsmaybedivided intoparametricandnonparametricvarieties,whichwillbecoveredinthefollowingsubsections.

1.2IntelligentAutomationInitiative

Intelligentautomationinitiative(IAI)isanemergingtechnology-drivenapproach tooptimizebusinessprocessesanddecision-makingthroughacombinationofAI, roboticprocessautomation(RPA),andotheradvancedtechnologies.Itaimsto streamlinerepetitiveandmundanetasks,improveproductivity,reduceerrors, andenableemployeestofocusonhigher-value-addedactivities.TheIAIstrategy involvestheintegrationofdifferenttechnologiestoautomatevariousaspectsof thebusiness,includingcustomerservice,supplychainmanagement,finance, humanresources,andmore.ThemaincomponentsofIAIinclude:

• Artificialintelligence(AI):Asubsetofcomputersciencethatfocusesondevelopingalgorithmsthatcanmimichumanintelligence,suchasspeechrecognition,NLP,ML,andcomputervision.AIhelpsorganizationstomakesenseof vastamountsofdata,predicttrends,andmakeinformeddecisions.

• Roboticprocessautomation(RPA):Asoftwaretoolthatusesbotstoautomate repetitiveandrule-basedtasks,suchasdataentry,invoiceprocessing,andreport generation.RPAcanreduceoperationalcosts,improveaccuracy,andincrease efficiency.

• Advancedanalytics:Itinvolvestheuseofstatisticalmodels,datamining,and predictiveanalyticstoanalyzedataandextractinsights.Thiscanhelporganizationstomakeinformeddecisionsandimprovebusinessoutcomes.

• Chatbots:AI-poweredvirtualassistantsthatcaninteractwithcustomers, answerqueries,andresolveissuesinrealtime.Chatbotscanimprovecustomer satisfaction,reduceresponsetimes,andfreeupresourcesforothertasks.

• Machinelearning:AsubsetofAIthatfocusesondevelopingalgorithmsthatcan learnfromdatawithoutbeingexplicitlyprogrammed.MLcanbeusedtomake predictions,identifypatterns,andautomatedecision-making.

• Cognitiveautomation:InvolvestheuseofAIandotheradvancedtechnologies toautomatecomplextasksthatrequirehuman-likereasoninganddecisionmaking.Thiscanincludetaskssuchasfrauddetection,riskanalysis,andsupply chainoptimization.

1.2.1BenefitsofIAI

IAIisastrategicapproachtointegratingadvancedtechnologies,suchasAI,RPA, andML,toautomatebusinessprocessesandworkflows.Herearesomeofthe benefitsofimplementingIAI:

• Increasedproductivity:Automationcanperformrepetitiveandtime-consuming tasksfasterandwithfewererrorsthanhumans,leadingtoincreasedproductivityandefficiency.Byfreeingupemployeesfromthesemundanetasks,theycan focusonhigher-valuetasksthatrequirecreativityandcriticalthinking.

• Costsavings:Automationcanhelpreducelaborcosts,ascompaniesnolonger needtohireadditionalstafftoperformrepetitivetasks.Additionally,automationcanhelpreduceoperationalcostsbystreamliningprocessesandreducing thepotentialforerrorsanddelays.

• Improvedaccuracyandquality:Automationcanperformtaskswithahigh degreeofaccuracy,consistency,andquality,reducingthepotentialforerrors andimprovingthequalityofworkproduced.

• Fasterprocessingtimes:Automationcanhelpspeedupprocessingtimesfor taskssuchasdataentry,dataanalysis,andreportgeneration.Thiscanlead tofasterdecision-makingandimprovedbusinessagility.

• Enhancedcustomerexperience:Automationcanhelpimprovethecustomer experiencebyenablingfasterresponsetimestoinquiries,reducingerrors, andprovidingmoreaccurateandpersonalizedservices.

• Increasedscalability:Automationcanhelpbusinessesscaletheiroperations moreeasilybyenablingthemtohandlehighervolumesofworkwithoutthe needforadditionalstaff.

• Betterdatainsights:Automationcanhelpbusinessesgatherandanalyzedata morequicklyandaccurately,enablingthemtomakebetter-informeddecisions.

Overall,thebenefitsofIAIcanhelpbusinessesstreamlinetheiroperations, reducecosts,andimprovetheirabilitytocompeteinanincreasinglyfast-paced andcompetitivemarket.

1.3ArtificialIntelligenceVersusIntelligentAutomation

AIandintelligentautomation(IA)aretworelatedtechnologiesthataretransformingthewaybusinessesoperate.AIisthesimulationofhumanintelligenceprocessesbymachines,whileIAreferstotheautomationofprocessesusingAI andotheradvancedtechnologies.IAcombinesRPA,ML,andotherAItechnologiestoautomaterepetitiveandtime-consumingtasks.Itallowsbusinessesto

automateprocessesthatwerepreviouslydonemanually,whichsavestime, reducescosts,andimprovesaccuracy.

AI,ontheotherhand,isabroaderfieldthatencompassesarangeoftechnologies,includingML,NLP,andcomputervision.Thesetechnologiesenable machinestoperformtasksthatwouldtypicallyrequirehumanintelligence,such asunderstandinglanguage,recognizingimages,andmakingdecisionsbasedon data.WhenAIandIAarecombined,businessescanachieveevengreaterbenefits. Forexample,IAcanbeusedtoautomateprocessessuchasdataentryanddocumentprocessing,whileAIcanbeusedtoanalyzethatdataandprovideinsightsfor decision-making.Thiscanhelpbusinessesmakemoreinformeddecisionsfaster, whichcanleadtoimprovedefficiency,productivity,andprofitability.Moreover, AIcanhelpautomatedecision-makingprocessesbyanalyzingvastamountsof dataandprovidingrecommendationsbasedonthatdata.IAcanthenbeused toexecutethosedecisionsautomatically,furtherstreamliningbusinessprocesses. ThecompleteworkflowofhowIAworksisshowninFigure1.1.

1.3.1ProcessDiscovery

Processdiscoveryinvolvesusingmathematicalequationsandalgorithmstoanalyzebusinessprocessesandidentifyareaswhereautomationcanbeapplied.One exampleofamathematicalequationusedinprocessdiscoveryistheprocesscycle efficiency(PCE),asshowninEquation(1.4).

PCE = VT CT ×100 ,

wherevalue-addedtime(VT)isthetimespentonactivitiesthatdirectlyaddvalue tothecustomer,andcycletime(CT)isthetotaltimetakentocompletetheprocess, includingbothvalue-addedandnon-value-addedactivities.The PCE formula helpsbusinessesidentifyareaswherethereiswastageorinefficiencyintheprocess.Ahigh PCE indicatesthataprocessishighlyefficientandthatthereisminimal wastage,whilealow PCE suggeststhatthereisalotofwastagethatcanbeeliminatedthroughautomation.

Thefirststepinprocessdiscoveryistocollectdataonthecurrentbusiness processes.Thiscanbedonebyconductinginterviewswithkeystakeholders, analyzingdocumentation suchasprocessmaps,orob servingtheprocesses

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.