Consider the following scenario. When organizations build machine-learning models, an assumption is made that the model will be accurate as long as the data patterns the model has been trained on remains true. If new data patterns emerge, or if the model has not been trained on all possible data sets or workflows, the model might be biased and provide inaccurate results. “A solution to this problem is to apply closed-loop machine-learning updates (meaning Edge AI) by detecting accuracy-level degradations and triggering automatic retraining,” explains Malladi. “Live data is then pushed back to a retraining module for model updates and subsequently, a new model is pushed back down to the edge to bring accuracy levels back to the original state. In summary, Edge AI helps organizations analyze live data streams and deliver intelligence at or near the source, leading to increased overall productivity, efficiency, and cost savings.”
Chetan Sharma, CEO of Chetan Sharma Consulting, says the edge market is just getting started, but it has been accelerating throughout 2020. “It is like the new gold rush, where everyone wants to own a piece of the business,” Sharma says. “There are many more participants in the value chain than just the cloud players, so the ecosystem will be more interesting and diverse.”
Sharma says edge access will become hugely important for latency-sensitive IoT services. The market, he adds, is being driven by industries like industrial and entertainment. Video analytics and AR (augmented reality) are other key drivers. The ability to push processing to the edge, rather than shipping all the data to the cloud for analytics offers huge bandwidth savings. To the list of market drivers, Malladi adds industries like green tech, automotive, and smart buildings. In green tech, efforts to take environmental responsibility and minimize carbon footprints are driving edge-computing implementations. “For example, fleet transportation organizations are beginning to deploy edge-based sustainability efforts to detect abnormal regen and idling events in realtime—ultimately reducing billions of pounds of CO2 emissions per year,” Malladi says.