機器學習簡介OutlineWhatismachinelearning(ML)TypesofmachinelearningWorkflowPopularmodelsApplicationsFuturesWhatismachinelearningTrainingset(labelsknown)Testset(labelsunknown)f()=“apple”f()=“tomato”f()=“cow”WhatismachinelearningDefinitionMachinelearningreferstoasystemcapableoftheautonomousacquisitionandintegrationofknowledgeMachinelearningisprogrammingcomputerstooptimizeaperformancecriterionusingexampledataorpastexperienceComputerDataAlgorithmProgramKnowledgeKnowledge(new)WhatismachinelearningEverymachinelearningalgorithmhasthreecomponentsRepresentationModel(rules,statistics,instance;logic,KNN,SVM,DNN,…)EvaluationPerformance(accuracy,mse,energy,entropy,…)OptimizationParametersCombinatorialoptimizationConvexoptimizationConstrainedoptimizationTypesofmachinelearningSupervisedlearningTrainingdataincludesdesiredoutputsUnsupervisedlearningTrainingdatadoesnotincludedesiredoutputsSemi-supervisedlearningTrainingdataincludesafewdesiredoutputsReinforcementlearningRewardsfromsequenceofactionsTypesofmachinelearningSupervisedlearningClassification:discreteoutputRegression:continuousoutputBias-variance大家應(yīng)該也有點累了，稍作休息大家有疑問的，可以詢問和交流8TrainingandValidationDataFullDataSetTrainingDataValidationDataIdea:traineachmodelonthe“trainingdata”andthentesteachmodel’saccuracyonthevalidationdataUnderfitting&OverfittingPredictiveErrorModelComplexityErroronTrainingDataErroronTestDataIdealRangeforModelComplexityOverfittingUnderfittingTypesofmachinelearningUnsupervisedlearningClusteringDimensionalityreductionFactoranalysisTypesofmachinelearningSemi-supervisedlearningClusteringorclassificationTypesofmachinelearningReinforcementlearningRobot&controlWorkflowPredictionTrainingLabelsTrainingTrainingImageFeaturesImageFeaturesTestingTestImageLearnedmodelLearnedmodelSlidecredit:D.HoiemandL.LazebnikWorkflowFeaturesWorkflowModelsLogic,RulesStatistical,BlackboxmodelStatic,dynamicmodelOnlinelearningEnsemblelearningWorkflowArchitectureModelFeatureHardwarePopularmodelsLinearmodel:logisticregression,lineardiscriminantanalysis,linearregression(withbasisfunction)PopularmodelsNearestneighborFeature&distancePopularmodelsSupportvectormachinePopularmodelsArtificialneuralnetworkPopularmodelsDecisiontreePopularmodelsCollaborativefilteringPopularmodelsHierarchicalclusteringK-meansSpectralclusteringManifoldlearningPopularmodelsHiddenmarkovmodelConditionalrandomfieldsApplicationsApplicationsApplicationsApplicationsApplicationsApplicationsApplicationsApplicationsApplicationsAttentionApplicationsImageclassificationApplicationsApplicationsBrainmachineinterfaceApplicationsApplicationsApplicationsApplicationsApplication