IoT revolution is ended. Thanks to hardware improvement, building an intelligent ecosystem is easier than never before for both startups and large-scale enterprises. The real challenge is now to connect, process, store and analyze data: in the cloud, but also, at the edge. We’ll give a quick look on frameworks that aggregate dispersed devices data into a single global optimized system allowing to improve operational efficiency, to predict maintenance, to track asset in real-time, to secure cloud-connected devices and much more.
12. Devices
Fleet
Edge
Techs
Tracking
Neural Network
Deep Learning
CPU / GPU / TPU
...at the Edge...
● ML inference at the Edge
● Dedicated TPUTM and VPU
● Local compute
● Run on Android Things and on various GNU/Linux-based OS
● Support enhanced security through hardware-root-of-trust
● Securely connected devices to Cloud providers
● Works seamlessly with cloud for hassle-free device provisioning
● Serverless scalability
15. ...and in the Cloud
Fetch
Clean
Prepare
Train Model
Evaluate
Model
Monitor /
Prediction /
Projection
Tune
Model
Parameters
Deploy
Updated
Model
★ continuous
deploy/optimization of
training model
★ update remotely OTA in
real-time
★ on selected group of
devices
Abstract: IoT revolution is ended. Thanks to hardware improvement, building an intelligent ecosystem is easier than never before for both startups and large-scale enterprises. The real challenge is now to connect, process, store and analyze data: in the cloud, but also, at the edge. We’ll give a quick look on frameworks that aggregate dispersed devices data into a single global optimized system allowing to improve operational efficiency, to predict maintenance, to track asset in real-time, to secure cloud-connected devices and much more.
Nice to meet you!
In the abstract for this session I said something really strong: the IoT revolution is ended. Wohw. He’s mad, you probably thought. But IoT is now already part of our everyday life. We're surrounded by this technology every day.
We are floating in globally dispersed devices. IoT is already a fact.
So today we're not gonna be talking about the Internet of Things, that sounds pretty generic: we’ll talk about data; in particular about “Deliver Data”.
Data allow us to deliver real-time predictive alert to our customers, to hunt for a parking spots we don't have to circle around the block three or four times or to optimize our energy consumption everyday/everywhere.
Analysts expect IoT devices to reach 16 millions units by 2020,
And What they do? They are generating more and more data day by day.
Petabyte. Exabyte every day!
Ok.
The mission here is to deliver data efficiently.
Obviously, you are doing that in the cloud: your personal cloud, Microsoft Azure, Amazon AWS or Google Cloud, it doesn’t matter at the moment.
So, at the end, this is your typical scenario. In 2019.
The Cloud Infrastructure has its physical layer, composed simply by Storage, a lot of CPUs and network switches. And by a service layer: composed by the various applications-as-a-service, infrastructure-as-a-service, platform-as-a-service, cloud functions, dataflow, data processing, etc…
But now the problem is “How to scale in the future with this upcoming massive epic gargantuesque amount of data?”.
Data Ingestion and Processing could become tricky.
The easier reply is: act directly on the physical layer of the cloud infrastructure. Doesn’t matter is you are using scalable cloud functions, fully-managed cloud services or whatever… at the end, in the real world, you are...
...requesting more and more resources to the cloud provider. More space, more bandwidth, upscaling the processing needs.
And, this what does it mean?
Moneys, a lot of moneys spent! More resources is equal to adding more expenses. So, are we sure is this the best way to scale?
Bring intelligence directly on the IoT devices at the Edge, that are connected to the Cloud.
In addition:
Discover how efficiently your devices operate, manage global assets, and carry out firmware updates on Cloud IoT platform, Predictive maintenance, Real-time asset tracking, Logistics & supply chain management (fleet management, inventory tracking, cargo integrity monitoring, and other business-critical functions), Build a comprehensive solution that spans across billions of sensors and edge devices, and bring a new level of intelligence and automation to entire homes, buildings, or cities.
Securely connect a few or millions of your globally dispersed devices through protocol endpoints that use automatic load balancing and horizontal scaling to ensure smooth data ingestion under any condition.
This devices allow onboard ML inference and supports multiple different onboard frameworks, including TensorFlow Lite, NN API, Caffe, ...
Finally, the hardware returns to have a major role!
So, this is our updated scenario since the introduction of AI-enabled IoT device at the Edge. In 2019.
But it’s not ended here. There is an amazing additional feature we can take advantage of...
But, the cloud remain critical even using this “Intelligence at the Edge” approach.
1) Acquire Data.
2) Clean the data to optimize the quantity and quality. Just select only the data that really matter for your scenario. Remove Noise. And simply trash the rest.
3) Prepare the data compressing in smaller bundle.
4) Train the model using the new new data.
5) Evaluate the model.
6) Make your projection/prediction.
7) Tune the model hyperparameters locally on your computer.
8) Deploy the refined model on the device at the edge.
This can be done in a continuous looped approach. Keeping the Edge device always optimized remotely.
Resuming
Alternative Scenario
If you are interested in going deepen into the topic, and if you want to get your hand dirty on this technology, you can find all the TPU and VPU development kit on Mouser.com.