机器学习简介ppt

-MachineLearning：AnOverview石立臣-Outline•Whatismachinelearning(ML)•Typesofmachinelearning•Workflow•Popularmodels•Applications•Futures-WhatismachinelearningTrainingset(labelsknown)Testset(labelsunknown)f()=“apple”f()=“tomato”f()=“cow”-Whatismachinelearning•Definition–Machinelearningreferstoasystemcapableoftheautonomousacquisitionandintegrationofknowledge–MachinelearningisprogrammingcomputerstooptimizeaperformancecriterionusingexampledataorpastexperienceComputerDataAlgorithmProgramKnowledgeKnowledge(new)-Whatismachinelearning•Everymachinelearningalgorithmhasthreecomponents–Representation•Model(rules,statistics,instance;logic,KNN,SVM,DNN,…)–Evaluation•Performance(accuracy,mse,energy,entropy,…)–Optimization•Parameters–Combinatorialoptimization–Convexoptimization–Constrainedoptimization-Typesofmachinelearning•Supervisedlearning–Trainingdataincludesdesiredoutputs•Unsupervisedlearning–Trainingdatadoesnotincludedesiredoutputs•Semi-supervisedlearning–Trainingdataincludesafewdesiredoutputs•Reinforcementlearning–Rewardsfromsequenceofactions-Typesofmachinelearning•Supervisedlearning–Classification:discreteoutput–Regression:continuousoutputBias-variance-TrainingandValidationDataFullDataSetTrainingDataValidationDataIdea:traineachmodelonthe“trainingdata”andthentesteachmodel’saccuracyonthevalidationdata-Underfitting&OverfittingPredictiveErrorModelComplexityErroronTrainingDataErroronTestDataIdealRangeforModelComplexityOverfittingUnderfitting-Typesofmachinelearning•Unsupervisedlearning–Clustering–Dimensionalityreduction–Factoranalysis-Typesofmachinelearning•Semi-supervisedlearning–Clusteringorclassification-Typesofmachinelearning•Reinforcementlearning–Robot&control-WorkflowPredictionTrainingLabelsTrainingTrainingImageFeaturesImageFeaturesTestingTestImageLearnedmodelLearnedmodelSlidecredit:D.HoiemandL.Lazebnik-Workflow•Features-Workflow•Models–Logic,Rules–Statistical,Blackboxmodel•Static,dynamicmodel•Onlinelearning•Ensemblelearning-Workflow•ArchitectureModelFeatureHardware-Popularmodels•Linearmodel:logisticregression,lineardiscriminantanalysis,linearregression(withbasisfunction)-Popularmodels•Nearestneighbor–Feature&distance-Popularmodels•Supportvectormachine-Popularmodels•Artificialneuralnetwork-Popularmodels•Decisiontree-Popularmodels•Collaborativefiltering-Popularmodels•Hierarchicalclustering•K-means•Spectralclustering•Manifoldlearning-Popularmodels•Hiddenmarkovmodel•Conditionalrandomfields-Applications-Applications-Applications-Applications-Applications-Applications-Applications-Applications-Applications•Attention-Applications•Imageclassification-Applications-Applications•Brainmachineinterface-Applications-Applications-Applications-Applications-Applications•Indirectillumination–Regression-Applications•Indirectillumination–kd-tree-Applications•Thecoreisthedataset!!!•Others–features–model&optimization-Futures•Decision•Control•Knowledge•Prediction