Adding Intelligence to the Edge Devices with Cloud IoT - Devfest FL

Transcript

1. Adding Intelligence to the Edge
   Devices with Cloud IoT
   DEVFEST FL
   NOV 16, 2019
   

   View Slide

2. Agenda
   Cloud IoT Introduction
   Adding Intelligence to the Edge Devices
   Coral Dev board & USB Accelerator
   Demo
   

   View Slide

3. HELLO!
   I’m SivamuthuKumar
   Architect - Computer Enterprises Inc
   Orlando, FL
   #Cloud #Mobile #IoT
   ksivamuthu
   

   View Slide

4. IoT
   INTERNET OF THINGS
   

   View Slide

5. IoT is the concept of
   connecting any device
   to the Internet and to
   other connected
   devices
   

   View Slide

6. View Slide

7. View Slide

8. It is predicted that there will be 41.6 billion
   connected IoT devices, or "things," generating
   79.4 zettabytes (ZB) of data in 2025
   This Photo by Unknown Author is licensed under CC BY-NC-ND
   

   View Slide

9. Cloud IoT Core
   DATA INGESTION AND DEVICE
   MANAGEMENT
   

   View Slide

10. View Slide

11. Cloud IoT Core
    Bi-directional communication with billions of IoT devices
    ◦ Device-to-cloud telemetry data, cloud-to-device command, track message delivery
    Work with familiar platform and protocols
    ◦ HTTP, MQTT protocols and clients
    Security Enhanced Solutions
    ◦ Individual identities and credentials for each of connected devices.
    Automate device provisioning to accelerate IoT deployment
    ◦ Register and provision devices with zero touches, in a highly secure and scalable way.
    Logging and Monitoring
    ◦ Audit logs and device logs
    

    View Slide

12. Cloud ML
    Fully managed service
    Preprocess and Orchestrate ML Workflow as
    Dataflow pipeline
    Analyze data and develop ML models in Datalab
    Training at scale using TensorFlow
    Batch and online predictions using REST at scale
    

    View Slide

13. ML Processing Units
    

    View Slide

14. View Slide

15. Cloud IoT Edge
    

    View Slide

16. Why ML at Edge?
    LATENCY BANDWIDTH PRIVACY CONNECTIVITY EXECUTION
    COST
    

    View Slide

17. View Slide

18. Edge TPU
    GOOGLE’S PURPOSE-BUILT ASIC DESIGNED TO RUN INFERENCE
    AT THE EDGE.
    

    View Slide

19. Edge TPU Development Kits
    Google Coral Dev Board USB Accelerator
    An individual Edge TPU is capable of performing 4 trillion operations (tera-operations) per second (TOPS),
    using 0.5 watts for each TOPS (2 TOPS per watt).
    

    View Slide

20. Demo
    RASPBERRY PI4 AND CORAL
    USB ACCELERATOR
    

    View Slide

21. Donuts
    This Photo by Unknown Author is licensed under CC BY-SA
    This Photo by Unknown Author is licensed under CC BY-SA
    

    View Slide

22. Cloud IoT Edge –Three parts
    TFLite Models
    ( Compatible with Edge TPU )
    Hardware
    Coral Dev board / USB Accelerator
    Edge TPU Runtime
    

    View Slide

23. Cloud AutoML Vision
    

    View Slide

24. Hardware
    ØAny Linux computer x86-64 or ARM64 system architecture
    ØOne available USB Port
    ØGoogle’s Coral USB Accelerator
    

    View Slide

25. Edge TPU Runtime
    

    View Slide

26. Edge TPU Delegates
    Install the TensorFlow Lite runtime Library
    Run the python Interpreter API with experimental Edge delegates.
    TF Lite API or Edge TPU API:
    

    View Slide

27. Boat
    Classification
    

    View Slide

28. Object
    Detection
    

    View Slide

29. View Slide

30. View Slide

31. Key
    Takeaways
    High speed ML inference using Edge TPU on low
    power devices.
    Cloud AutoML to train and export into TFLite and
    TFLite Edge optimized models.
    Retrain / Transfer Learning using AutoML or on your
    own and compile into Edge TPU TFLite models
    Edge TPU with Coral SoM or USB Accelerator (
    USB3.0 & Host computing)
    

    View Slide

32. Reference
    Demo:
    https://github.com/ksivamuthu/cloud-edge-tpu-demo
    Slides:
    https://speakerdeck.com/ksivamuthu/adding-intelligence-
    to-the-edge-devices-with-cloud-iot
    Cloud IoT Core:
    https://cloud.google.com/iot-core/
    Coral Dev Kit & Accelerator:
    https://coral.withgoogle.com/products/accelerator/
    

    View Slide

33. View Slide

34. Thank you
    ksivamuthu
    

    View Slide