Autonomous cars need to identify road signs in real time, drones need to recognize objects with or without network connectivity. In this breakout session, you will learn what is machine learning (ML) inference at the edge and why it matters. We will show you how to use AWS Greengrass to locate cloud trained machine learning models, deploy them to your Greengrass devices, enable access to on-device GPU or FPGA, and apply the models to locally generated data without a need for connection to the cloud.
11. 12
Machine Learning in SONY Factories
Already running Greengrass on our factory floors, now developing ML for:
Factory operator positioning
• Operator location & time resource management
• Using beacon and ML predict position of operator
Predictive maintenance
• Detect aging degradation of bearing with acceleration sensor
Size limitation (ML library is big)
Tight coupling ML model and Lambda
HW resource access
Needed more AWS Lambda ability
12. 13
Using Greengrass ML Inference
Deploy and run customized ML model on devices using Greengrass for “Anomaly Detection”
Customize model without size limitation
Continuous adjustment , delivery model
HW resource access
(high resolution sensor)
SONY: Spritzer
Host board (GG Core)
Accelerometer
Manufacturing
Machine
Special device + host
(Accelerometer)
SONY ML dashboard
Customize model
for various factory machine
Easy to customize model
Management cloud
Detection
acceleration
x,y,z
Class label
Cloud
E d g e
Greengrass ML Inference
Sony now integrates “Cloud, Edge, Firmware and Sensor Devices”
31. 33
AWS Greengrass & Intel – Powerful,
Optimized edge solutions
Delivers a secure, intelligent edge
Developers can easily create new applications,
from edge to cloud
32. AWS Greengrass & Intel- Machine learning
capabilities at the edge
34
Intel Atom Processor with integrated graphics
Deep Learning Optimized Software
Compute performance, agility, and speed
to run real time machine learning on the device.