Applications of Neural Networks - Tutorialspoint

Why Artificial Neural Networks? · With the help of neural networks, we can find the solution of such problems for which algorithmic method is expensive or does ... ArtificialNeuralNetworkTutorial ArtificialNeuralNetwork-Home BasicConcepts BuildingBlocks Learning&Adaptation SupervisedLearning UnsupervisedLearning LearningVectorQuantization AdaptiveResonanceTheory KohonenSelf-OrganizingFeatureMaps AssociateMemoryNetwork HopfieldNetworks BoltzmannMachine Brain-State-in-a-BoxNetwork OptimizationUsingHopfieldNetwork OtherOptimizationTechniques GeneticAlgorithm ApplicationsofNeuralNetworks ArtificialNeuralNetworkResources QuickGuide UsefulResources Discussion SelectedReading UPSCIASExamsNotes Developer'sBestPractices QuestionsandAnswers EffectiveResumeWriting HRInterviewQuestions ComputerGlossary WhoisWho ApplicationsofNeuralNetworks Advertisements PreviousPage NextPage  BeforestudyingthefieldswhereANNhasbeenusedextensively,weneedtounderstandwhyANNwouldbethepreferredchoiceofapplication. WhyArtificialNeuralNetworks? Weneedtounderstandtheanswertotheabovequestionwithanexampleofahumanbeing.Asachild,weusedtolearnthethingswiththehelpofourelders,whichincludesourparentsorteachers.Thenlaterbyself-learningorpracticewekeeplearningthroughoutourlife.Scientistsandresearchersarealsomakingthemachineintelligent,justlikeahumanbeing,andANNplaysaveryimportantroleinthesameduetothefollowingreasons− Withthehelpofneuralnetworks,wecanfindthesolutionofsuchproblemsforwhichalgorithmicmethodisexpensiveordoesnotexist. Neuralnetworkscanlearnbyexample,hencewedonotneedtoprogramitatmuchextent. Neuralnetworkshavetheaccuracyandsignificantlyfastspeedthanconventionalspeed. AreasofApplication Followingsaresomeoftheareas,whereANNisbeingused.ItsuggeststhatANNhasaninterdisciplinaryapproachinitsdevelopmentandapplications. SpeechRecognition Speechoccupiesaprominentroleinhuman-humaninteraction.Therefore,itisnaturalforpeopletoexpectspeechinterfaceswithcomputers.Inthepresentera,forcommunicationwithmachines,humansstillneedsophisticatedlanguageswhicharedifficulttolearnanduse.Toeasethiscommunicationbarrier,asimplesolutioncouldbe,communicationinaspokenlanguagethatispossibleforthemachinetounderstand. Greatprogresshasbeenmadeinthisfield,however,stillsuchkindsofsystemsarefacingtheproblemoflimitedvocabularyorgrammaralongwiththeissueofretrainingofthesystemfordifferentspeakersindifferentconditions.ANNisplayingamajorroleinthisarea.FollowingANNshavebeenusedforspeechrecognition− Multilayernetworks Multilayernetworkswithrecurrentconnections Kohonenself-organizingfeaturemap ThemostusefulnetworkforthisisKohonenSelf-Organizingfeaturemap,whichhasitsinputasshortsegmentsofthespeechwaveform.Itwillmapthesamekindofphonemesastheoutputarray,calledfeatureextractiontechnique.Afterextractingthefeatures,withthehelpofsomeacousticmodelsasback-endprocessing,itwillrecognizetheutterance. CharacterRecognition ItisaninterestingproblemwhichfallsunderthegeneralareaofPatternRecognition.Manyneuralnetworkshavebeendevelopedforautomaticrecognitionofhandwrittencharacters,eitherlettersordigits.FollowingaresomeANNswhichhavebeenusedforcharacterrecognition− MultilayerneuralnetworkssuchasBackpropagationneuralnetworks. Neocognitron Thoughback-propagationneuralnetworkshaveseveralhiddenlayers,thepatternofconnectionfromonelayertothenextislocalized.Similarly,neocognitronalsohasseveralhiddenlayersanditstrainingisdonelayerbylayerforsuchkindofapplications. SignatureVerificationApplication Signaturesareoneofthemostusefulwaystoauthorizeandauthenticateapersoninlegaltransactions.Signatureverificationtechniqueisanon-visionbasedtechnique. Forthisapplication,thefirstapproachistoextractthefeatureorratherthegeometricalfeaturesetrepresentingthesignature.Withthesefeaturesets,wehavetotraintheneuralnetworksusinganefficientneuralnetworkalgorithm.Thistrainedneuralnetworkwillclassifythesignatureasbeinggenuineorforgedundertheverificationstage. HumanFaceRecognition Itisoneofthebiometricmethodstoidentifythegivenface.Itisatypicaltaskbecauseofthecharacterizationof“non-face”images.However,ifaneuralnetworkiswelltrained,thenitcanbedividedintotwoclassesnamelyimageshavingfacesandimagesthatdonothavefaces. First,alltheinputimagesmustbepreprocessed.Then,thedimensionalityofthatimagemustbereduced.And,atlastitmustbeclassifiedusingneuralnetworktrainingalgorithm.Followingneuralnetworksareusedfortrainingpurposeswithpreprocessedimage− Fully-connectedmultilayerfeed-forwardneuralnetworktrainedwiththehelpofback-propagationalgorithm. Fordimensionalityreduction,PrincipalComponentAnalysis(PCA)isused. UsefulVideoCourses Video ProloginArtificialIntelligence 78Lectures 7hours ArnabChakraborty MoreDetail Video ArtificialintelligenceinJavascriptGamedevelopment-TicTacToeAI 87Lectures 9.5hours DigiFisk(ProgrammingIsFun) MoreDetail Video IntroductiontoArtificialIntelligence:AIforbeginners MostPopular 10Lectures 1hours NikolozSanakoevi MoreDetail Video ArtificialIntelligence:TheFutureOfProgramming 15Lectures 54mins MukundKumarMishra MoreDetail Video ArtificialIntelligenceLevel1:Cogito 11Lectures 1hours GiladJames,PhD MoreDetail Video ArtificialIntelligenceLevel2:RiseoftheMachines 20Lectures 2hours GiladJames,PhD MoreDetail PreviousPage PrintPage NextPage  Advertisements Print  AddNotes  Bookmarkthispage  ReportError  Suggestions Save Close Dashboard Logout