April 21, 2021

Edge AI: Why we built it.

By Arjun Ramamurthy, Head of Video Product, O&O, Verizon Media, and Debashis Mondal, Principal Product Manager, Verizon Media

For new and emerging technologies to gain broad acceptance, they must be capable of unlocking new business efficiencies and bring value to consumers. Both individually and in combination, AI, big data, cloud computing and IoT have proven their value. These technologies help businesses capture huge amounts of data streaming from up to millions of sources, store it in massive cloud data centers and use machine learning (ML) and artificial intelligence (AI) techniques to derive valuable, game-changing insights in near real time.

At Verizon Media, we started our big data and artificial intelligence journey in 2014. We built a cloud-based platform capable of ingesting petabytes of data and used machine learning to deliver near real-time insights to improve the performance of our delivery network, enhance efficiency across our network and deliver an even better customer experience. We then built additional AI and big data applications for customers in order to, for example, gain insights into customer behavior by using wireless usage data.

In many ways, these technological advancements are already revolutionizing industries, and we’re only at the very beginning of a long and exciting road ahead. But with big data, cloud-based AI and IoT, many of the most powerful and value-producing applications have not been implemented due to processing and latency delays in the cloud, storage costs, and security and privacy concerns. And with AI, it’s not acceptable for answers to come back in days, minutes, or seconds. For many applications, intelligence needs to be applied in the moment, within milliseconds.

The most obvious and compelling answer to overcoming these obstacles is to move processing closer to the end device or the source of data generation—to the edge of the network. Verizon is an industry leader in edge computing with a CDN that offers massive network capacity and just 10-25 milliseconds of latency from virtually every internet user on the planet. For workloads that require even less latency, our 5G network offers <10 ms latency. The emergence of edge computing combined with AI, IoT and big data processing opens the door to entirely new services and consumer value. By placing intelligence at the edge, breakthrough AI applications can now function in near real time.

Starting with Verizon’s global edge computing capabilities, merged with our AI know-how, we have designed and built a fully integrated platform we call Edge AI, on top of which we are developing specific vertical applications. This is the first platform of its kind, with all the building blocks for end-to-end Edge AI solutions already in place.

The opportunity for increased value unlocked by moving AI applications to the edge is massive. It presents a huge opportunity for network and solution providers who have the  capabilities to bring robust Edge AI solutions to market. In many ways, it is a perfect confluence for our service offering, allowing us to build on existing strengths and create a foundation for building an intelligent, secure, extensible and reliable platform capable of accommodating new services, use cases and applications.

In a recent report titled “5G and AI: The Foundation for the Next Societal and Business Leap,” ABI Research predicts that the low latency provided by edge computing and AI in combination are “likely to transform the way we live and work.” The report goes on to say these technologies will “pave the way for a variety of new business opportunities in the consumer and enterprise segments, otherwise not possible with existing technologies.” ABI Research estimates that AI and ML applications deployed at the edge will create $3.1 trillion worth of value by 2025. And as the technologies reach maturity, they forecast value creation worth 9.2% of the global gross domestic product (GDP) in 2035. As shown in the chart below, this massive economic impact comes primarily through productivity gains made possible when these technologies are used in combination.

Figure 1: Within 15 years, edge computing will have a significant impact on GDP.

The growing importance of edge computing

For more than the past decade, organizations have collected data from IoT devices and sensors or visual and audio recordings from cameras and microphones deployed across their facilities before transporting it to a centralized data center or cloud for further analysis and storage.

There are multiple problems with this approach as the number of IoT devices continues to ramp up. Tech analyst firm IDC projects there will be 55.7 billion connected IoT devices by 2025. They note these devices will generate 73.1 zettabytes (ZB) of data, a significant uptick from 18.3 ZB in 2019. IDC argues this growth will require organizations to rethink long-term data storage strategies and look for opportunities in analytics/AI at the edge.

Let’s consider an industrial or manufacturing enterprise where you have many thousands of sensors. As the number of sensors increases, it’s simply not practical to send the vast amounts of data flowing from these sensors to the cloud, have the analytics done there, and then send the results back to the manufacturing location to eventually act on the insights from that data.There are multiple challenges to this process, including:

  1. Sending all the data to the cloud requires huge amounts of bandwidth 
  2. Storing everything in the cloud exponentially increases your cloud storage costs
  3. Moving certain kinds of sensitive information to the cloud puts that data at risk 

These types of operational efficiency issues are all resolved, or at least significantly minimized, through edge computing, whether it be servers running nearby as part of a CDN or on-premises via 5G mobile edge computing (MEC) infrastructure for public or private networks.

Finally, one of Edge AI’s most important aspects is its ability to provide low latency for real-time use cases. By placing the processing capability close to end devices, Edge AI dramatically reduces the lag that can occur between data ingestion/acquisition, processing and the action required at the end. Cutting latency is critical to enabling innovative applications, spanning everything from connected vehicles and more immersive gaming and media experiences to more intelligent and fast-paced manufacturing environments.

Edge AI also opens the opportunity to use connected IoT devices and ML applications in environments where reliable internet/Wi-Fi connectivity (or any at all) may not be a given, such as a deep sea drilling rig or research vessel, or airports. An intelligent application for monitoring environmental conditions, such as the presence of hazardous gases, would be useless if it was dependent on cloud connectivity.

Edge AI applications

Edge AI has the potential to be the technological platform for many new applications across a wide range of industries and use cases. For instance, in the manufacturing segment, Edge AI will enable the widespread implementation of Industry 4.0 initiatives, including more predictive analytics, automated factory floors, reconfigurable production lines and optimized logistics. Media and entertainment can use Edge AI to localize content and increase personalization. Edge AI can drive across-the-board improvements to urban infrastructures by enabling more advanced applications in the education/public sector, telemedicine or by taking transportation automation to the next level. The possibilities are nearly endless.

To showcase the capabilities and versatility of our Edge AI platform, we have developed a number of prototypes and demonstrations, including a:

  • Predictive failure detection solution
  • Smart shelf system for retail using computer vision 
  • Stranger detection system for a lab or manufacturing facility 
  • Automated optical inspection system for quality assurance

For the predictive failure detection application, sensors are mounted on motors and other equipment and configured to continually stream signals on temperature, vibration and current to the Edge AI platform. Instead of sending all data to the cloud, the AI analyzes the data continuously locally to make predictions for when a particular motor is about to fail. By accurately detecting anomalies and failure conditions in advance, a plant or maintenance supervisor can take corrective actions to prevent a production outage. And since data storage and analysis occur within the plant location, organizations gain more timely alerts coupled with improved data security and reduced data storage and bandwidth costs.

Figure 2. Predictive failure analysis driven by Edge AI can minimize the risks of unpredicted failures and outages.

In the working prototype for an automated retail inventory tracking system, an autonomous shelf scanning robot delivers camera feeds to an ML model running on the Edge AI platform. The system can recognize objects and deliver inventory details to a dashboard, providing more frequent, accurate and comprehensive insights on inventory status along with real-time status monitoring and alerting for low inventory.

Figure 3. Automated retail inventory tracking powered by Edge AI provides insights for monitoring trends and delivers real-time alerts.

Despite electronic locks and other security measures, it can be challenging to prevent unauthorized access to restricted areas in sprawling manufacturing facilities or lab environments. By incorporating video feeds from across the facility into the Edge AI, the system can use facial recognition to detect strangers and provide real-time notification to security personnel.

Figure 4. Edge AI improves security by using computer vision to identify strangers in a manufacturing or lab environment.

Another use case for Edge AI and computer vision is automated optical inspection on manufacturing lines. In this case, assembled components are sent through an inspection station for automated visual analysis. The Edge AI computer vision model detects missing or misaligned parts or any other flaw and delivers results to a real-time dashboard showing inspection status. Because the data can flow back into the cloud for further analysis, the ML models can be continuously improved to reduce false positives. By increasing the speed and accuracy of defect identification, the system improves manufacturing yield and increases process throughput.

As these examples illustrate, the integration of AI and compute capacity combined with cloud services directly into the network’s edge enables organizations to bring increasingly sophisticated and transformative use cases to market. As a fully integrated platform, Edge AI significantly reduces the hurdles involved with bringing these use cases to life.

In the next installment of what will be a three-part series, we will take a deeper dive into the Edge AI platform’s elements.

Contact us to learn more about how your application could benefit from our Edge AI platform.

Contact us

Sales

Support

Manage your account or get tools and information.

More info