Edge Computing and Machine Learning: Two Sides of the Same Coin
by Brad Johnson, on Nov 21, 2017 11:33:39 AM
The pace of adoption for Edge Computing technologies in industrial environments is picking up. Industry leaders value the potential benefits of analytics efforts, which can inform better maintenance strategies, improve productivity, and reduce wastage. However, in order to achieve these benefits, businesses must process massive volumes of data generated at the edge. While it is possible to push all data to the cloud for analysis, dealing with data at the edge enables applications to run more efficiently, while reducing burden on networks and centralized databases.
A major difference between Cloud and Edge Computing infrastructures involves how real-time data in treated in edge environments. Traditional cloud architectures focus moving data to the cloud, where it can be processed at a later time. This is primarily a logistical challenge. Architects set up buffers and other mechanisms in order to throttle data flow into manageable streams. They may even deploy memory resources at the edge, in order to facilitate for extreme buffer lengths and to ensure quality of service. However, the actual processing occurs in the cloud, far away from the data source and after having incurred significant latency penalties.
Edge Computing Treats Data Differently
Edge Computing flips this model on its head. Data is processed as close to the source as possible, while higher order computations (for example, aggregate statistics) are performed at whichever level in the application hierarchy makes the most sense (from an efficiency perspective). In theory, only the highest level computations would then occur in the cloud. But in practice, the entire edge-based application functions as a single cohesive cloud. The difference is that the edge-based architectures optimize for proximity, and therefore minimize incurred latency.
Because sensor data is processed on/near the physical edge devices, applications can access structured, real-time data streams within the edge network with minimal latency. These high rate edge streams are ideal for performing Machine Learning (ML) at the edge, and can inform predictive maintenance and other IT/OT applications. Data processed at the edge is made accessible sooner to operators, or can create a real-time feedback loop with automation systems. Time series data retains the critical element of “timeliness,” meaning that proactive measures can be informed by Machine Learning prediction models in real-time, compared with post hoc batch data analysis which could be stale by minutes or hours after central processing.
Machine Learning at the Edge
Lastly, Edge Computing architectures allow for a more efficient distribution of compute resources. Looking at the example of traffic intersections, Chris Sachs, a founder and Lead Architect of Swim, explains that “it's roughly a trillion times more expensive to train a single network on 100 intersections than it is to train 100 networks on overlapping groups of 20 intersections.” Edge computing provides a greater surface area for computation to occur. By ensuring that computations occur as close to the source as possible, application complexity can be measurably reduced.
These complexity and latency benefits accumulate to magnify impact and effectiveness of Machine Learning efforts. Machine Learning applications benefit from the “freshness” of real-time edge data, with all time-context retained because data is made available in real-time. This allows for more frequent and higher resolution intervals of observation, while additional context can lead to more accurate predictions. It’s possible to build cloud-based Machine Learning applications that use static, batch analytics technologies to generate stale predictions. But if the goal is to create forward-looking, continuously updating Machine Learning applications, which predict future states based real-time data, then Edge Computing and Machine Learning are two sides of the same coin.
Learn how SWIM uses Edge Computing to deliver real-time edge data insights with millisecond latency for industrial and other real-time applications.