PROBABILISTIC POTENTIALFUNCTION NEURALNETWORK CLASSIFIER - - PDF document

▶
probabilistic potential function neural network classifier
SMART_READER_LITE
LIVE PREVIEW

PROBABILISTIC POTENTIALFUNCTION NEURALNETWORK CLASSIFIER - - PDF document

Apr 22, 2023 405 likes •515 views

PROBABILISTIC POTENTIALFUNCTION NEURALNETWORK CLASSIFIER GurselSerpen LloydG.Allred* KrzysztofJ.Cios

slide-1
SLIDE 1
  • PROBABILISTIC

POTENTIALFUNCTION NEURALNETWORK CLASSIFIER

  • GurselSerpen

LloydG.Allred* KrzysztofJ.Cios

  • ElectricalEngineering&ComputerScienceDepartment

TheUniversityofToledo,Toledo,OH43606

  • *SoftwareEngineeringDivision

OgdenAirLogisticsCenter HillAFB,UT84056

slide-2
SLIDE 2

DesiredPropertiesofProposed Classifier

  • ♦Real-timetrainingandclassification

♦Multi-modallydistributedclasses ♦Classes formed from a set of disconnected subclasses ♦Noinitialguessforthenetworktopology ♦Discoverclusteringpropertiesoftrainingdata ♦Adapttoaminimalnetworktopology

  • ♦Implementincrementallearningprocedure

♦Form optimal decision boundaries - a theoreticalBayesianclassifier.

slide-3
SLIDE 3

SignificantNeuralClassification Paradigms

  • ♦ TheMulti-LayerFeedforwardNetwork

Initialnetworktopology-needsguessing slow training speed - unsuitable for real-time implementations.

  • ♦ TheRadialBasisFunctionNetwork

Initializationofnetwork-clusteringpropertiesof trainingdata(k-means) Hiddenlayernodecount

  • ♦ TheProbabilisticNeuralNetwork

A pattern layer node for each training pattern Potentiallylargenodecounts

  • ♦ LearningVectorQuantizationNetworks

Codebookvectorinitialization-nowell-defined procedureexists

slide-4
SLIDE 4

TopologyofPPFNN.

  • MAXNET

Layer Hidden Layer Pattern Layer Output Layer I N P U T S

wij

O U T P U T S

slide-5
SLIDE 5
  • NETWORKCREATION
  • PROCESS

2-CLASS PROBLEM

slide-6
SLIDE 6
  • LVQ

MLP PPFN N 20 40 60 80 100 LVQ MLP PPFN N Classification algorithms

Classificationratein%

  • 2-SpiralData
  • LVQ

RBF MLP PPFNN 10 20 30 40 50 60 70 80 90 100 LVQ RBF MLP PPFNN Classificationalgorithms

Classificationratein%

  • IrisTestData
slide-7
SLIDE 7

LVQ RBF MLP PNN PPFNN 10 20 30 40 50 60 70 80 LVQ RBF MLP PNN PPFNN Classificationalgorithms

Classificationratein%

  • SonarTestData
  • LVQ

RBF MLP PPFNN 10 20 30 40 50 60 LVQ RBF MLP PPFNN Classificationalgorithms

Classificationratein%

  • VowelTestData
slide-8
SLIDE 8

CONCLUSIONS

  • PerformanceofProposedAlgorithm
  • Initialfindingsareverypromising

BenchmarkProblemstestedinclude2-Spiral, IRIS,Vowel,andSonar

  • SimulationResults
  • Fasttrainingandclassification

Highclassificationrateonproblemstested Minimalnetworksize Smallnumberofempiricallydetermined parameters Roomtoimproveperformance

slide-9
SLIDE 9
  • CURRENTWORK
  • Morebenchmarkproblems

Determineawaytodefineoptimalvaluesfor potentialfunctionspread Determineanoptimalsequenceforweight values Improveclassificationperformance-optimality