Transformation to Artificial Intelligence with MATLAB Roy Lurie, PhD Vice President of Engineering MATLAB Products

Transcription

1 Transformation to Artificial Intelligence with MATLAB Roy Lurie, PhD Vice President of Engineering MATLAB Products 2018 The MathWorks, Inc. 1

2 A brief history of the automobile First Commercial Gas Car 1885 Benz Patent Motor Car 2

3 A brief history of the automobile First Affordable Car 1908 Ford Model T 3

4 A brief history of the automobile First ECU 1981 GM Vehicles 4

5 A brief history of the automobile 1981 TODAY 5

6 The First Digital Transformation: Embedded Controls in Everything? Model-Based Design Data Analysis Engine Calibration 6

7 A brief history of the automobile First Connected Cars 1998 BMW Assist Safety & Convenience Services 7

8 A brief history of the automobile First ADAS 2000 Mercedes-Benz Actros Lane Departure Warning System 8

9 A brief history of the automobile 2000 Today/Soon 9

10 The Second Digital Transformation: Artificial Intelligence Everywhere? Model-Based Design Data-Driven Algorithms 10

11 11

12 Using Machine Learning to build and deploy a predictive maintenance system 1TB Pump logs of temperature, pressure & other data Analytics and Machine Learning plus signal processing, neural networks & more Predictive Model deployed to drill site Maintenance Needed 12

13 13

14 Using Model-Based Design and Data Analytics to design and verify an automated driving system 80 TB Machine Learning extract interesting events Visualize Sensor Data Vehicle Logs of video and radar data Sensor Fusion Algorithms deployed on vehicle Re-simulate Refine Algorithm 14

15 Challenges of developing and integrating AI Big Data Analytics Machine Learning and Deep Learning Cloud and Embedded Algorithm Deployment Making AI easy and accessible for Engineers 15

16 Making big data easy and accessible for Engineers Working with Real-World Data Numeric double, single, logical categorical datetime duration calendarduration Heterogeneous timetable structure cell table Text c h char {c h} cell string str string 16

17 Making big data easy and accessible for Engineers Working with Really Big Data {c h} {c h} {c h} {c h} {c h} {c h} str str str str str str Tall arrays 17

18 Making machine learning easy and accessible for Engineers Machine Learning Apps Regression Learner 18

19 Deep Learning is Changing Expectations 20

20 21

21 Deep Learning is Changing Expectations 22

22 23

23 Deep Learning Tools are for Programmers 24

24 Making deep learning easy and accessible for Engineers 25

25 Making deep learning easy and accessible for Engineers camera = webcam; img = snapshot(camera); net = alexnet; label = classify(net,img) 26

26 Making deep learning easy and accessible for Engineers Automate ground truth labeling using deep learning 28

27 Sometimes you need more than object detection 29

28 Making deep learning easy and accessible for Engineers Labeling everything through semantic segmentation 30

29 Delivering analytics to both embedded and enterprise systems ENTERPRISE EMBEDDED 31

30 Making deep learning easy and accessible for Engineers Automatically generate CUDA code and TensorRT to deploy trained deep learning networks to NVIDIA GPUs 32

31 Making deep learning easy and accessible for Engineers Embedded GPU code finding drivable path 33

32 Each release makes AI easy and accessible JSON Hyperparameter Optimization MATLAB Drive CAFFE Model Zoo Mobile Sensors Cloud Logging Spark Heatmap Charts Tall timetable Timetable Machine Learning Code Gen Regression Learner app Add-On Explorer Categories Missing Data Functions Tall Arrays Deep Learning MATLAB Production Server RESTful Interface AlexNet Classification Code Gen Deep Learning on CPU VGG-16, VGG-19 Categorical Plotting Bayesian Statistics Symbolic Variable Units GPU Coder Tall Visualization Live Functions Image Segmenter App DAG Networks LSTM Semantic Image Segmentation Image Denoising Deep Learning Training Visualization Text Analytics Toolbox Change Point Detection Twitter Datafeed MQTT Geographic Bubble Chart Parallel Bayesian Optimization Web Apps App Testing Framework uidatepicker Faster Startup Faster Image and Video Reading Deep Learning Layer Validation More Tall Array Statistics Modal Analysis Predictive Maintenance Toolbox 36

33 Unique Platform Combination Powering New Applications Control Systems Verification & Validation Model-Based Design Code Generation Big Data Machine Learning Data Analytics Signal and Image Processing 37

34 38