“The Data-Driven Engineering Revolution,” a Presentation from Edge Impulse

“The Data-Driven Engineering Revolution,” a Presentation from Edge Impulse

• 1.
  © 2021 EdgeImpulse The Data-driven Engineering Revolution Zach Shelby, Co-founder and CEO May 25, 2021
• 2.
  Copyright © EdgeImpulseInc.
• 3.
  © 2021 EdgeImpulse Embedded ML Makes Devices Smarter 3 Sensor Audio Vision Pattern Detected 85% Status Device From Cortex-M to Jetson Nano Starting at < 10 kB of Flash/RAM Real-time, Low-power Inference
• 4.
  © 2021 EdgeImpulse Embedded Programming Today 4 Rule-based Code Write More Code Trial and Error
• 5.
  © 2021 EdgeImpulse Data-driven Engineering Is Here! 5 Machine Learning Sensor Collect More Data
• 6.
  © 2021 EdgeImpulse 6 Input (data) Rules (model) + = Machine Learning Outcome (label) Machine Learning Finds The Rules Input Rules + = Outcome Programming
• 7.
  © 2021 EdgeImpulse Predictive Maintenance ● Industrial ● White goods ● Infrastructure ● Automotive Asset Tracking & Monitoring Human & Animal Sensing Motion, Current, Audio, Temp and Camera ● Logistics ● Infrastructure ● Buildings Motion, Temp, Humidity, Position, Audio and Camera ● Health ● Consumer ● Industrial Camera, Motion, Radar, Audio, PPG, ECG, Camera
• 8.
  © 2021 EdgeImpulse Embedded ML Makes Devices Smarter 8 Jetson Nano Cortex-A + GPU Video Classification Sensor Audio Vision Arduino Nano Cortex-M4 Voice Keywords Audio Classification 50 kB Arduino Pro Cortex-M7 Image Classification 250 kB+ Raspberry Pi Cortex-A Object Detection Complex Voice Processing 1 MB+ Cortex-M0+ Anomaly Detection Sensor Classification 20 kB
• 9.
  © 2021 EdgeImpulse Enabling ML For All Developers 9 Learn more at http://edgeimpulse.com/ ML lifecycle from sensor to deployment
• 10.
  © 2021 EdgeImpulse Solving The Entire ML And DevOps Lifecycle 10
• 11.
  © 2021 EdgeImpulse Edge Impulse For Linux • Less than 5 minutes from data collection to custom ML deployment • Announcing full support for the Raspberry Pi 3 and 4, NVIDIA Jetson and compatible Linux targets • Full native hardware acceleration for CPU and GPU • Data collection from cameras, microphones, and any custom sensor over GPIO • Integrate with a few lines of code: Python, Node.js, Go or C++ 11 edge-impulse-linux edge-impulse-linux-runner
• 12.
  © 2021 EdgeImpulse 12 Outcomes NB-IoT with 20+ year battery life made possible by ML Avoid disastrous wildfires and reduce maintenance costs Enable automated monitoring of poles and lines Electric grid solution to enable fault detection in real time with Edge Impulse Making The Grid Smarter With ML Power line fire detected! Status Motion Current Testimonial from IRNAS: “The efficiency gain in this project is from 1 man year for 3 FTE (embedded dev, data scientist and ML expert man month for 1 FTE by 1 subject matter expert who can do it all by himself”
• 13.
  © 2021 EdgeImpulse 1 3
• 14.
  © 2021 EdgeImpulse IoT Data! = ML Data 14 50 Billion Devices 4.4 Zettabytes Database Warehouse Lake
• 15.
  © 2021 EdgeImpulse Deploy ML Where It Makes Sense 15 Social Media ERP SCADA
• 16.
  © 2021 EdgeImpulse https://www.coursera.org/edgeimpulse
• 17.
  © 2021 EdgeImpulse Further Resources 17 2021 Embedded Vision Summit Demos • Low-compute Image Classification with Himax Grayscale Camera • Effortless Digit Recognition with Arduino Portenta • Power Up with Jetson Nano Object Detection • Build an Image Classification Solution Under $5 Panel Discussion • Can You Make Production ML Work Without Dozens of PhDs? Over-the-Shoulder Tutorial • Build Industrial Embedded ML Solutions on CPUs and MCUs Coursera Course on Embedded ML Smart Grid Case Study Get Started with Edge Impulse for Linux