Integrating Edge AI

By on
Edge AI

Integrating edge artificial intelligence (AI) is not a simple process. Early forms of artificial intelligence relied on the computer power of data centers to perform their processor-demanding tasks. After some time, AI shifted into software, using predictive algorithms that changed how these systems support businesses. AI has now moved to the outer edges of networks.

Artificial intelligence at the edge exists when local “edge” devices process AI algorithms instead of being processed in the cloud.

AI at the edge is reducing the restrictions and limitations of applications using deep neural networks. To operate efficiently, edge AI applications should have high-speed and low-power processing, and include advanced integrations designed specifically for the tasks needing to be accomplished.

For example, consider an edge AI system using vision inputs that can use a single camera to provide quality control for a production line, or multiple cameras for the safety of a self-driving car or a mobile robot.

Jason Mann, the vice president of the IoT at SAS, explained:

“Edge AI happens when AI techniques are embedded in internet of things (IoT) endpoints, gateways, and other devices at the point of use.”

Edge AI Planning

Intelligent planning is key to developing and integrating an edge AI system. Creating a list of needs and desires (such as which equipment should be monitored) is a good start. This can be followed by creating a map of where the sensors and endpoints are located.

Artificial intelligence and machine learning have traditionally relied heavily on the resources of the cloud to support their processes. The cloud can be used to train deep neural networks. Once a framework for a deep neural network is created, it can be installed at the edge for use.

Use of AI at the edge allows for real-time operations that include data creation, data storage, and decision-making. 

At this point, installing a deep learning algorithm at the edge allows for real-time analysis of high-volume data, in just a few milliseconds. Once the system is in place, edge AI lowers internet bandwidth requirements, in turn lowering costs by minimizing the amounts of data being transferred back and forth to the cloud. Intelligent questions to ask as part of the planning phase include:    

  • What type of data, and how much, will be collected and analyzed? Data volume and storage needs should be assessed.
  • What is involved in making the endpoints secure? Security at the endpoints is a significant consideration.
  • What is the maturity level of the AI technology being considered? A major consideration is whether there is a pretrained or prebuilt AI model available that fits the organization’s needs.
  • What is needed to support the desired edge AI algorithms? Determine the technology needed for deploying AI on the edge.

Bring in the Experts and Set Up a Pilot Project

Upgrading to an edge AI system is expensive, and not something you want to pay for twice. Creating a small scale template, a mini-version of the internet of things project, is an excellent way to create a test run and identify the problems that can develop.

At this point, bringing in experts or AI consultants can be a very valuable step. Finding the experts can be an adventure. They may not exist locally. You can search online for edge AI consultants to find a “remote” consultant.

IBM has a free course titled Edge AI engineer, which could be used to train inhouse staff. Having staff trained on, and familiar with, your edge AI system presents some interesting advantages (a knowledgeable staff, faster than average repair times).

Combining a remote edge AI consultant and trained inhouse staff would be very functional.

Once the organization is ready in terms of equipment and planning, then it is time to begin installing and integrating. Start small and have the project’s goals in mind. The pilot project should not take longer than three months. (Much of this time will be spent waiting for a problem to surface.)

Remember to include storage as part of the plan. Storing data sent from the sensor to the edge device allows for finding “patterns” in the data. These patterns may come from the data sent by medical devices (the internet of medical things or IoMT) or it may be sent by a sensor in a factory (the industrial internet of things or IIoT). The patterns may prevent health issues from arising or provide clues for preventative maintenance.

Selecting and Integrating the AI Edge Devices

AI at the edge uses algorithms that are processed on local edge devices, instead of being processed in the cloud. AI decisions are made more quickly because the data from the sensors doesn’t have to travel far. This technology reduces the limitations of automotive and industrial applications.

Developing faster, smarter, and more efficient systems require more data and more sensors to supply the data.

This means increasing the amount of processing power is also necessary. It should be noted these increases may cause problems in a computer system’s performance — this should be considered in the planning phase. When designing the edge AI system, it is important to recognize the limitations of the current computer system.

It is also important to realize edge devices are basically interfaces. They will need to support common interface technologies (USB, Ethernet, CAN, serial, and/or GPIO). They will also have to support peripheral equipment, such as displays, cameras, and keyboards.

The environment where the edge device is located is also a concern. An edge device may be placed in a location with extremes in humidity, temperature, or vibrations. This consideration should affect which device is selected, and how it is housed or packaged.

Another consideration is regulatory requirements. A device using radio frequencies is subject to regulations. Some devices will comply as they come “out of the box.” Others may need some additional efforts to make them compliant.

Choosing the right solution requires a careful assessment of the business’ needs. Does the edge device need to be on constantly, or will it be asleep for long stretches of time? Is it triggered by an external event such as detected motion, lights coming on, a switch being flipped)?

Some Edge AI Device Sources

  • Advian offers edge AI devices for manufacturing, mining, forestry, and chemical industries. They also provide edge AI devices for financial institutions, retail, and the energy industry. (And they have consultants!)
  • Nvidia offers edge AI devices for healthcare, retail, electric cars, etc. (You may want to contact Nvidia to see if they have the devices you need—or can create them.)
  • AAEON eShop  The BOXER-8521AI supports Edge AI Computing per the Google Edge TPU System.
  • IBM offers edge computing (and consultants).

Edge AI Computing Steps into the Future

Edge AI provides faster computing, better security, and more efficient control while supporting continuous operations. It also supports improved machine learning and advanced algorithms. Many global businesses have begun taking advantage of edge AI. The benefits range from improving the monitoring of assembly lines to self-driving vehicles. Edge AI can benefit a variety of industries.

With the recent roll out of 5G technology, edge AI has gained an added boost. Some of the applications benefitting from edge AI are:

  • Automated optical inspection has become an important part of manufacturing. It can detect faulty parts on a production line using an automated edge AI optical analysis, without relying on large amounts of communication with the cloud.
  • Virtual assistants, such as Alexa or Siri have benefitted from edge AI, which allows ML algorithms to deep learn more quickly using data stored on the edge device, rather than the cloud. (Theoretically, this should offer a more individualized virtual assistant.)
  • Self-driving vehicles, with the help of edge AI, can identify objects in the road more quickly, and make faster decisions. This results in both faster and safer transportation.

Image used under license from

Leave a Reply