Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron Fundamentals of Artificial Neural Networks () May22,2009 1/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron Outline Introduction ABriefHistory FeaturesofANNs NeuralNetworkTopologies ActivationFunctions LearningParadigms FundamentalsofANNs McCulloch-PittsModel Perceptron Adaline(AdaptiveLinearNeuron) Madaline CaseStudy: BinaryClassificationUsingPerceptron () May22,2009 2/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron Introduction ArtificialNeuralNetworks(ANNs)arephysicalcellularsystems,which canacquire,storeandutilizeexperientialknowledge. ANNsareasetofparallelanddistributedcomputationalelements classifiedaccordingtotopologies,learningparadigmsandattheway informationflowswithinthenetwork. ANNsaregenerallycharacterizedbytheir: Architecture Learningparadigm Activationfunctions () May22,2009 3/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron Typical Representation of a Feedforward ANN () May22,2009 4/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron Interconnections Between Neurons () May22,2009 5/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron History A Brief History ANNshavebeenoriginally designedintheearlyfortiesforpattern classificationpurposes. ⇒Theyhaveevolvedsomuchsincethen. ANNsarenowusedinalmosteverydisciplineofscienceandtechnology: fromStockMarketPredictiontothedesignofSpaceStationframe, frommedicaldiagnosistodataminingandknowledgediscovery, fromchaospredictiontocontrolofnuclearplants. () May22,2009 6/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron Features of ANNs ANNareclassifiedaccordingtothefollowing: Architecture ActivationFunctions LearningParadigms Feedforward Binary Supervised Recurrent Continuous Unsupervised Hybrid () May22,2009 7/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron NeuralNetworkTopologies Neural Network Topologies FeedforwardFlowofInformation () May22,2009 8/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron NeuralNetworkTopologies Neural Network Topologies (cont.) RecurrentFlowofInformation () May22,2009 9/61 Introduction Features Fundamentals Madaline CaseStudy:BinaryClassificationUsingPerceptron ActivationFunctions Binary Activation Functions StepFunction SignumFunction 1, ifx >0 1, ifx >0 step(x)= sigum(x)= 0, ifx =0 0, otherwise (cid:26) −1, otherwise 2 2 1 1 0 0 -1 -1 -2 -2 0 2 -2 -2 0 2 () May22,2009 10/61
Description: