WIXLIB

1

센서 데이터 애널리틱스를 위한신호처리 및 머신러닝 기법이웅재 부장Application Engineering GroupRyan.Lee@mathworks.com 2016 The MathWorks, Inc.2

Sensor Analytics and Edge Node Development3

Key topics§Signal Processing methods (e.g. digital filtering, spectral analysis)§Machine Learning algorithms (e.g. classification)§MATLAB environment “enablers” (e.g. language, visualization, Apps,documentation)§Flow from predictive algorithms to embedded implementation(e.g. DSP system simulation, automatic code generation)4

Sensor Analytics WorkflowAcquireData Acquisition ToolboxInstrument Control ToolboxHardware SupportPackages (XX)Database ToolboxThingSpeakAnalysisSimulinkSignal Processing ToolboxDSP System ToolboxAnalyticsMATLABStats & Machine Learning TBNeural Networks ToolboxDeploymentMATLAB CoderSimulink CoderEmbedded CoderMATLAB CompilerMATLAB Production Server5

Signal analysis for classificationApplication examples§Mobile sensing§Structural health monitoring (SHM)§Fault and event detection§Automated trading§Radar post-processing§Advanced surveillance§.6

Example: Human Activity Analysis and et courtesy of:Davide Anguita, Alessandro Ghio, Luca Oneto, Xavier Parra and Jorge L. Reyes-Ortiz.Human Activity Recognition on Smartphones using a Multiclass Hardware-Friendly Support Vector Machine.International Workshop of Ambient Assisted Living (IWAAL 2012). Vitoria-Gasteiz, Spain. Dec 2012http://archive.ics.uci.edu/ml/datasets/Human Activity Recognition Using Smartphones7

Sensor Data Analytics Workflow – the bigger pictureSignalProcessingMachineLearning Domain knowledgeOpen-ended problemLong discovery cycles Built-in algorithmsConcise code (54 lines for 66 features!)Apps and visualisation accelerate insight8

Sensor Data Analytics Workflow – the bigger pictureConnect andAcquireSignalProcessing Different tools and environmentsDisconnect between hardware and analysisInefficiencies in data sharing MATLAB Connects to DAQ interfaces andsensors directly. E.g. Android Sensor Support iPhone and iPad Sensor SupportMachineLearning9

Sensor Data Analytics Workflow – the bigger pictureConnect andAcquireMachineLearningSignalProcessing EmbeddedImplementationSignal analysis vs. on-line DSPFrom Machine Learning theory to pretrained, low-footprint classifiersMATLAB vs. C/C Streaming algorithms, data sources andvisualization for System modelling and simulationAutomatic code generation10

Leverage Built-in Algorithms, Apps, and Technologies§Signal Processing Toolbox Built-in algorithms and Apps to process andanalyse ogramxcovfindpeaks 11

Leverage Built-in Algorithms, Apps, and Technologies§Signal Processing Toolbox §Parallel Computing Toolbox Accelerate computationally and data-intensiveproblems using multicore processors, GPUsand computer clusters§parfor12

Leverage Built-in Algorithms, Apps, and Technologies§Signal Processing Toolbox §Parallel Computing Toolbox §Statistics and Machine Learning Toolbox Functions and apps to describe, analyze, andmodel data.Regression, clustering and classificationalgorithms to draw inferences from data andbuild predictive models classificationLearner13

Leverage Built-in Algorithms, Apps, and Technologies§Signal Processing Toolbox §§§Parallel Computing Toolbox §Statistics and Machine Learning Toolbox §Neural Network Toolbox Functions and apps to design, train,visualize, and simulate neural networks nprtoolpatternnet14

Leverage Built-in Algorithms, Apps, and Technologies§§Signal Processing Toolbox Parallel Computing Toolbox §§§§§Statistics Toolbox §Neural Network Toolbox DSP System Toolbox Streaming algorithms, data sources andvisualization for system modelling atorSpectrumEstimatorTimeScope15

Leverage Built-in Algorithms, Apps, and Technologies§Signal Processing Toolbox §Parallel Computing Toolbox §Statistics Toolbox §Neural Network Toolbox §DSP System Toolbox §MATLAB Coder Generate embeddable source C/C from MATLAB code codegen(Learn more: MATLAB to C Made Easy webinar)16

Signal Processing and Machine LearningTechniques for Sensor Data AnalyticsSummary§Extensive set of de-facto standard functions forsignal processing and machine learning§Environment accelerates insight and automation:visualisation, apps, language, documentation§Path to embedded products, from on-linesimulation to automatic code generation17

Neural Networks Toolbox MATLAB Coder Simulink Coder Embedded Coder MATLAB Compiler MATLAB Production Server Acquire Analysis Analytics Deployment. 6 Signal analysis for classification Application examples § Mobile sensing § Structural health monitoring (SHM) § Fault