How Machine Learning enables edge computing

This post is the transcript of an informational video I created. You can find the video at the bottom of the post.

With the maturing of IoT technology, we can cover the world in sensors, and give computers access to the physical world in a way that just wasn’t possible before.

Here’s an example. You have a factory, with lots of machines designed to assemble a widget. From time to time, one of these machines breaks, disrupting production and leaving you with a big headache. Having heard of “predictive maintenance”, you decide to place sensors all over the production line. These sensor collect data continuously.

The data collected by the sensors is very often not easily understood by humans. We just don’t deal very well with high-dimensional mathematics. Luckily, we have computers, and we have figured out many techniques to teach them to recognize patterns in high-dimensional, complex data. This set of techniques go under the name of “Machine Learning”.

With Machine Learning, experts can create automated systems that can alert you when something in the production line is not behaving as it should, allowing you to schedule timely repairs, saving money and preventing headaches.

But this is not the end of the story.

In a typical setup, the automated system lives on a server, and receives a continuous stream of raw data directly from the sensors on the production line. This comes at a very high cost, because of the sheer amount of data being continuously produced, moved, crunched and stored. A solution to this problem is to move intelligence from the central server towards the sensors, a paradigm known as “Edge computing”.

Instead of continuously sending raw data to a central powerful computer, we can put many small computers where the sensors live, and teach them to, in a way, “reduce the chatter”. That is, we can again use Machine Learning to make these small computers only send high-level, condensed information, and only when it matters.

The number of sensors in the world is about to grow from the order of billions to that of trillions. It is increasing much faster than our ability to deal with the data they generate. To fully harness the power of the huge sensor network we are building for our physical world, we must distribute intelligence in the network itself, make sensors smarter and able to decide what to communicate, when, and where to send the information.

If you have enjoyed this post, and you are looking for experts in the field of Machine Learning for the Internet of Things, check us out at foldAI, drop me an email, or contact me on LinkedIn