Machine And Deep Learning With MATLAB - MathWorks

Machine and Deep Learning with MATLABAlexander Diethert, Application EngineeringMay, 24th 2018, London 2018 The MathWorks, Inc.1

2

AgendaArtificial Intelligence enabled by Machine and Deep LearningMachine LearningDeep LearningOutlook: Integration in Production Systems3

Source:Gartner, Real Truth of Artificial Intelligence by Whit AndrewsPresented at Gartner Data & Analytics Summit 2018, March 20184

Big DataCompute PowerMachine LearningAnalytics are pervasive – Why Now?We have dataWe have computeWe know how EngineeringBusinessTransactionalDesktopMulticore, GPU Clusters Cloud computing Hadoop with SparkNeural NetworksClassificationClusteringRegression and much more 5

6

7

There are two ways to get a computer to do what you wantTraditional ProgrammingDataCOMPUTERProgramOutput

There are two ways to get a computer to do what you wantMachine LearningDataCOMPUTEROutputProgram

There are two ways to get a computer to do what you wantMachine LearningDataCOMPUTEROutputArtificial IntelligenceMachine LearningModel

AI, Machine Learning, and Deep LearningArtificialIntelligenceMachineLearningAny techniquethat enablesmachines tomimic humanintelligenceStatistical methodsenable machines to“learn” tasks fromdata without explicitlyprogrammingDeep LearningNeural networks with many layers thatlearn representations and tasks“directly” from dataDeep Learning moreaccurate than humans onimage adrillion11

What can Machine and Deep Learning do?http://www.cs.ubc.ca/ nando/340-2012/lectures/l1.pdf12

Example: Predictive Analytics in e-commerceUse Image ProcessingImagesEngineering DataSocial profileGeolocationKeystroke logsto add image data to the model,improving performanceIMPROVEDPredictiveModelImprovedOffer toCustomerBusiness DataTransactions13

Applications of Machine Learning and Deep Learning inFinanceAlgorithmic TradingSentiment AnalysisForecasting / PredictionFraud DetectionCredit Decision MakingFinancial Planning14

AgendaArtificial Intelligence enabled by Machine and Deep LearningMachine LearningDeep LearningOutlook: Integration in Production Systems15

Customer References16

Example: Machine Learning for Risk ManagersMachine learning is enabling better models for complex problemshttps://www.mckinsey.com/ /media/mckinsey/dotcom/client service/risk/pdfs/the future of bank risk management.ashx17

Machine Learning Workflow1324Integrate withProductionSystemsPreprocess DataDevelop PredictiveModelsFilesWorking withMessy DataModel Creation e.g.Machine LearningDesktop AppsDatabasesData prise ScaleSystemsAccess and Explore Data5Visualize Results3rd partydashboardsAWSKinesisWeb d Devicesand Hardware18

Types of Machine LearningType of LearningUnsupervisedLearningCategories of AlgorithmsClusteringOutput is the # of groups formed fromsimilar data. Find natural groups andpatterns from input data onlyGroup & interpretdata based onlyon input dataMachineLearningClassificationOutput is a choice between classes(True, False) (Red, Blue, Green)SupervisedLearningDevelop predictivemodel based on bothinput and output dataRegressionOutput is a prediction of the future state19

Workflows of Machine LearningIterate: apply model, TERS CLUSTERING PREPROCESSDATASUPERVISEDLEARNINGFILTERS LSupervisedLearningTESTDATAPREPROCESSDATAMODEL .Class,State, FILTERS 1. ACCESSPREDICTION2. EXPLORE AND DISCOVER3. SHARE20

Demo: Classification Learner App21

Machine Learning Apps for Classification and Regression Point and click interface – nocoding required Quickly evaluate, compare andselect regression models Export and share MATLAB codeor trained models22

Fine-tuning Model ParametersWhy?o Manual parameter selection istedious and may result insuboptimal performanceHyperparameter Tuning with Bayesian OptimizationPreviously tuning theseparameters was amanual processWhen?o When training a model with oneor more parameters thatinfluence the fitCapabilitieso Efficient comparted to standardoptimization techniques or gridsearcho Tightly integrated with fitfunction API with pre-definedoptimization problem (e.g.bounds)23

Building out your Machine Learning ToolAccess and Explore DataProcess Data and CreateFeatureBuild and Validate ModelsDeploy ModelReview Model24

AgendaArtificial Intelligence enabled by Machine and Deep LearningMachine LearningDeep LearningOutlook: Integration in Production Systems25

Machine learning vs deep learningDeep learning performs end-to-end learning by learning features, representations and tasks directlyfrom images, text and soundDeep learning algorithms also scale with data – traditional machine learning saturatesMachine LearningDeep Learning26

What is Deep Learning?27

Data Types for Deep LearningSignalTextImage28

Deep learning and neural networks Deep learning neural networks; Data flows through network in layersLayers provide transformation of dataInput LayerOutput LayerHidden Layers (n)29

Thinking about Layers Layers are like blocks– Stack on top of each other– Replace one block with adifferent one Each hidden layer processesthe information from theprevious layer30

Thinking about Layers Layers are like blocks– Stack them on top of each other– Replace one block with adifferent one Each hidden layer processesthe information from theprevious layerLayers can be ordered indifferent ways31

Convolutional neural networks Train “deep” neural networks on structured data (e.g. images, signals, text)Implements Feature Learning: Eliminates need for “hand crafted” featuresTraining using GPUs for performancecartruckvan bicycleInputConvolution ReLuPoolingConvolution ReLuFeature ication32

Convolutional Neural Networks (CNN)Output data CNN take a fixed size input and generate fixed-size outputs.Convolution puts the input images through a set of convolutional filters,each of which activates certain features from the input data.Input data 33

Another Network for Signals - LSTM LSTM Long Short Term Memory (Networks)– Signal, text, time-series data– Use previous data to predict new information I live in France. I speak .c0C1Ct34

Long Short-Term Memory (LSTM) LSTM are an extension of Recurrent Neural Networks. RNN can handle arbitrary input/output lengths. They have the capability to use the dependencies among inputs. LSTMs just like every other RNN connect through time. They are capableof preserving the long-term and short-term dependencies that occur withindata.35

Example: Algorithmic Trading36

Another Application: Sentiment Analysis with Twitter DataAccess TweetsDevelop ModelPreprocess TweetsClean-up TextApple's iPhone 8 toDrive 9.1% Increase inShipments Per IDChttps://t.co/n085F65upk AAPL GRMN GOOGPredict SentimentConvert to Buyapples iphone driveincrease shipments peridcIncreaseFraud37

Deep Learning on CPU, GPU, Multi-GPU and ClustersHOWSingleCPUTO TA R G E T ?Single CPUSingle GPUSingle CPU, Multiple GPUsOn-prem server withGPUsCloud GPUs(AWS)38

GPU Coder Automatically generates CUDA Code from MATLAB Code– can be used on NVIDIA GPUs CUDA extends C/C code with constructs for parallel computing39

AgendaArtificial Intelligence enabled by Machine and Deep LearningMachine LearningDeep LearningOutlook: Integration in Production Systems40

Integrate with Production SystemsDataAnalyticsDatabasesCosmos DBCloud StorageStreamingPI SystemMATLABProduction ServerDashboardsWebAzureBlobAWSKinesisBusiness SystemRequestBrokerCustom AppsAzureIoT HubPlatform41

Thank you for your attention42

Artificial Intelligence enabled by Machine and Deep Learning Machine Learning . Real Truth of Artificial Intelligence by Whit Andrews Presented at Gartner Data & Analytics Summit 2018, March 2018. 5 Big Data Compute Power Machine Learning . Data Analytics Business System MATLAB Produc