7
process decisions and identify operational problems. Devices can also use ML to
analyze local data locally rather than passing the data to a remote server or to the
cloud to extract intelligence.
For example, consider an application that initiates a process when a human being
is detected. That application may use an imaging sensor to collect data once every
second for analysis. Rather than firing up a radio to transmit that data to the cloud
and using cloud resources to analyze the images, a low-power local ML algorithm can
analyze each image, make a judgement about the probability of that image containing
something that looks human, and—if the probability is high enough—turning on the
process. "That approach provides a lot more flexibility, and it means that you don't have
to communicate with the cloud," Aronchick explains. "That doesn't mean you're not
communicating with the cloud. It's just that you're taking action based on
local information."
"That approach
provides a lot
more flexibility,
and it means
that you
don't have to
communicate
with the cloud."
