Transcription
The Rise of Engineering-DrivenAnalyticsRoy Lurie, Ph.D.Vice President Engineering,MATLAB Products 2015 The MathWorks, Inc.1
The Rise of Engineering-Driven Analytics2
The Rise of Engineering-Driven Analytics3
Limitedusers, scope& technologyBig DataMachineLearningComputePowerPervasiveusers, scope,& technologyAnalytics are now pervasiveDescriptive &Diagnostic PredictivePrescriptiveApply robust, statistically-motivated methods todata produced from complex systems toEngineering Desktop Neural Networksunderstand whathas happened,Multicore,GPUBusiness ClassificationTransactionalpredict ClustersCloudcomputingwhat willhappen,and Hadoop with Spark ClusteringRegression and much more suggest decisions or actions.4
Analytics in e-commerceUse Image ProcessingImagesEngineering DataSocial profileGeolocationKeystroke logsto add image data to the model,improving performanceIMPROVEDPredictiveModelOffer toCustomerBusiness DataTransactions5
Consider the Data in Data AnalyticsVideoAudioImagesEngineering DataSensorSocial profileGeolocationUsing nowKeystroke logsBusiness DataTransactionsLevel of Industry / User AdoptionSource: Gartner Big Data Industry Insights, March 20166
Consider the Data in Data AnalyticsVideoAudioPlannedImagesEngineering DataSensorSocial profileGeolocationUsing nowKeystroke logsBusiness DataTransactionsLevel of Industry / User AdoptionSource: Gartner Big Data Industry Insights, March 20167
The Rise of Engineering-Driven Analytics8
Architecture of an analytics systemData fromData from instrumentsbusinesssystemsand connected systemsAnalyticsand MachineLearning9
Architecture of an analytics systemData fromData from instrumentsbusinesssystemsand connected systemsMATLAB & Simulink Integrates inAnalyticsand MachineEmbedded Systems and EnterpriseLearningIT WorkflowsPredictive ModelPredictive Modeldeployed in smart andembedded systemsdeployed on cloud andbusiness systems10
11
25% costreduction12
Example – BuildingIQAdaptive building energy management13
Optimizing Energy Costs and Consumption at Building IQDATA - Billions of data points:MATLABToolboxes Just WorkPhysics, energy cost, power,internal temperatures,ambient temperatures, ambient humidity, buildingand work togetheroperation schedule, comfort bounds, etc.–!Analytics andMachine Learningplus system identification,control theory & moreCurrent energycosts & demandWeatherFeedsPredictive Modeldeployed on cloud with client systemand real-time data feeds14
Why MATLAB? MATLAB Impeccable Numerics forTrusted ResultsRobust numerical algorithmsExtensive visualization and analytics toolsIndustry-robust and reliable mathematicaloptimization routinesGood object-oriented frameworkAbility to interface with Java (for backend work)Running MATLAB in the cloud in productionUnit-testing frameworkWe could rapidly translate ourprototypes into productionalgorithms that deal reliablywith real-world noise anduncertaintyBorislav Savkovic, BuildingIQ15
Example – ScaniaAutomatic emergency braking using sensor fusion and analytics16
17
Using Model-Based Designto build and deploy the analyticsin an embedded control systemMATLAB Integrates Analytics andModel-Based Design18
Implementing Sensor Fusion at ScaniaMachine learningto develop fusion algorithmsfor situation detectionVehicle logsof video and radar dataPredictive Modeldeployed on vehicle19
AutomotiveOff-highway vehiclesAeronauticsThe Rise of Engineering-Driven AnalyticsRetailIndustrial AutomationFinanceOil & GasHealthcare managementMedical DevicesInternetClean Energy20
Predictive Maintenancefor polymer-based production machinesSensor Data ( 1 minute)10-100 sensors/machineQuality State ( 40 minutes)Classification using Statistics, MachineLearning, and Neural Networks21
Deployment – a MATLAB Appused by machine operatorsState NOT OKM153State OKM15722
23
The need for data scientistsDomainexpertiseCoding andintegration skillsStatistical andmathematicalknowledge24
What they say Expand university programs Train existing analysts25
26
TSMC Student Contestuse process control datato improve semiconductor yields 21 teams competedWafer Big Data in HadoopMATLAB used by winningteam and 2nd place team27
IoT open data platformfor students and makersBuilt-in MATLAB analysisSimulink support via Raspberry Pi28
MATLAB lets you be your own data scientistMATLAB is Designed andDocumented to be Easy forEngineers and Scientists to UseDomainexpertiseStatistical andmathematicalknowledgeCoding andintegration skills29
Limitedusers, scope,& technologyBig Data putePower DesktopMulticore, GPU Clusters Cloud computing Hadoop with Spark Pervasiveusers, scope,& technologyNeural NetworksClassificationClusteringRegression and much more In MATLABNative support forengineering data Database interfaces StreamingNEW for MATLABAudio System Toolbox R2016aVision HDL Toolbox R2015a30
Limitedusers, scope,& technologyIn MATLABBig Data EngineeringBusinessTransactional Native support forengineering dataDatabase interfacesStreamingDatastoretext, image, video,Excel filesTimetable, string, andtall arrays 2016b MachineLearningComputePower DesktopMulticore, GPU Clusters Cloud computing Hadoop with Spark Pervasiveusers, scope,& technologyNeural NetworksClassificationClusteringRegression31
Limitedusers, scope,& technologyIn MATLABBig Data EngineeringBusinessTransactional Native support forengineering dataDatabase interfacesStreamingDatastoretext, image, video,Excel filesTimetable, string, andtall arrays 2016b MachineLearningComputePower DesktopMulticore, GPU Clusters Cloud computing Hadoop with Spark Pervasiveusers, scope,& technologyNeuralNetworksis fast:MATLABClassification- heavily optimized librariesClustering- JIT compiledRegression- takes advantage of thecompute power you have Multicore & GPUMATLAB DistributedComputing Serverand EC2 Support Hadoop with Sparksupport R2016b MATLAB ProductionServer32
Limitedusers, scope,& technologyIn MATLABBig Data EngineeringBusinessTransactional Native support forengineering dataDatabase interfacesStreamingDatastoretext, image, video,Excel filesTimetable, string, andtall arrays 2016b MachineLearningComputePowerPervasiveusers, scope,& technology DesktopMulticore, GPU Clusters Cloud computing Hadoop with Spark Neural NetworksClassificationClusteringRegression Statistics and MachineLearning ToolboxClassification Learner App R2015aNeural Network ToolboxCNNs for Deep learning R2016aMachine learningwith code generationMulticore & GPUMATLAB DistributedComputing Serverand EC2 Support Hadoop with Sparksupport R2016b MATLAB ProductionServer 33
Classification Learner Appin Statistics and Machine Learning Toolbox34
Deep Learning with Neural Network Toolbox - New in R2016aTechnical Computing andData Analyticscamera webcam;img snapshot(camera);net alexnet;label classify(net,img)35
Example –First consumer otoscope in a mobile device usingmachine learning and computer vision36
The Rise of Engineering-Driven AnalyticsLimitedusers, scope,& technologyBig DataComputePowerMachineLearningPervasiveusers, scope,& technologyBe your own Data Scientist!37
4 Big Data Compute Power Machine Learning Limited users, scope & technology Pervasive users, scope, & technology Engineering Business Transactional Desktop - Multicore, GPU Clusters Cloud computing Hadoop with Spark Neural Networks Classification Clustering Regression and much more Analytics are now pervasive