Using machine learning to drive intelligence at the edge

Description

With 30+ billion connected things by 2020, the IoT will drive an explosion of data that will need to be processed, stored, managed, and analyzed—in most cases, in real time—to derive business value. However, the volume, diversity, speed, and inherent characteristics of data generated from sensors and the IoT will challenge traditional data management mechanisms. Given this, IoT architectures are increasingly focused on making the edge more intelligent in an effort to lower round-trip latencies and minimize cost of data transmission.

More importantly, machine learning is starting to play a key role in enabling IoT use cases today. Organizations need to be able to do advanced analytics to enable concepts such as pattern recognition, anomaly detection, and ultimately predictive modeling based on the petabytes of data that the IoT generates. And this is where large-scale machine learning and advanced analytics comes into play.

So how can organizations utilize machine learning, deep learning, and advanced analytics to make intelligent decisions closer to where data is generated? How can you build machine learning models for the IoT and push those out back to the edge? What role do open source technologies play in this end-to-end architecture?

The way to make the edge more intelligent is by building machine learning models in a centralized hub in the cloud and push the knowledge generated out back to the edge. The possibilities for use cases utilizing machine learning at the edge are endless.

Dave Shuman and Bryan Dean explore a recent proof of concept (PoC) that was executed at one of Europe’s leading global manufacturers. Cloudera, Red Hat, and Eurotech have deployed an end-to-end architecture, built on open source technologies, for the IoT that enables end-to-end analytics, including business rules and advanced analytical models that can be deployed both at the edge and within the centralized data platform. They were able to train a deep learning image recognition model of fire and deploy onto a constrained IoT edge gateway device, where it could evaluate if fire broke out in a factory floor in real time. They were able to demonstrate edge model execution for a machine learning model using Eclipse Deeplearning4j deployed as an OSGi bundle. Join Dave and Bryan to learn more about this project and discover how organizations are using advanced analytics and machine learning to make the edge more intelligent. Along the way, Dave and Bryan walk you through a roadmap illustrating how organizations can utilize machine learning and effectively push that knowledge back out to the edge. They then detail an end-to-end open source architecture for the IoT based on Eclipse Kura, an open source stack for gateways and the edge, and Eclipse Kapua, an open source IoT cloud platform. The architecture can enable:

• Capabilities to securely connect and manage millions of distributed IoT devices and gateways
• Machine learning and analytics capabilities with intelligence and analytics at the edge
• A centralized data management and analytics platform with the ability to build or refine machine learning models and push them out to the edge
• Application development, deployment, and integration services

They conclude with an Industry 4.0 demo that highlights how to ingest, process, and analyze data coming off of factory equipment and how to enable machine learning on the edge using all of this data.