[From the last episode: Home routersAn electronic box that helps steer data on a network. For instance, you may have one in your home connecting your phone and computer and other devices to each other and to the internet. The data itself has information about where it's being sent; the router uses that information to send it in the right direction. At a really basic level, you can think of a router and a switch as being the same thing. If you want to get more technical, a switch creates a local subnetwork, and the router connects multiple subnetworks (or multiple networks). and even devices can act as fog; in industry, often a dedicated gatewayA piece of electronic network equipment that takes a local network and gives it access to the internet. Your cable modem, for instance, might act as a gateway. is used.]

Before we leave our high-level section on computing notions, we have to address something that has many people excited – and many other people worried. Here I’m talking about artificial intelligenceA broad term for technology that acts more human-like than a typical machine, especially when it comes to "thinking." Machine learning is one approach to AI. (AI), and, more specifically, machine learningMachine learning (or ML) is a process by which machines can be trained to perform tasks that required humans before. It's based on analysis of lots of data, and it might affect how some IoT devices work. (ML). Technologists have talked about AI for decades, and approaches have come and gone. ML is newer, and is but one approach to AI – one that may have a longer life than past attempts.

Both AI and ML have been in the news lately, with the specific good/bad news characteristic that machinesIn our context, a machine is anything that isn't human (or living). That includes electronic equipment like computers and phones. can learn to do things that only humans used to be able to do. So humans have less tedious, menial work to do – which, according to prognosticators from years back, should free us up for more enriching activities. Except that our society rewards only work, and if work goes away without being replaced by other work (for all displaced workers, not just some of them), then we’re not kind to those left behind.

So, as you can see, machine learning goes way beyond the Internet of Things. It may be behind some of the lovingly (or not) referred-to algorithmsA way of calculating something, typically as a step-by-step recipe. A familiar one to all of us is how we do long division on paper: that's an algorithm. Algorithms abound in the IoT. that Google or Facebook or Amazon uses. But, with IoTThe Internet of Things. A broad term covering many different applications where "things"  are interconnected through the internet. gadgets, it’s one of the things that might happen with all of that data your devices sends into the cloudA generic phrase referring to large numbers of computers located somewhere far away and accessed over the internet. For the IoT, computing may be local, done in the same system or building, or in the cloud, with data shipped up to the cloud and then the result shipped back down..

Building a Model

The idea behind machine learning is to allow the machine to build a modelA simplified representation of something real. We can create models of things in our heads without even realizing we're doing it. Technology often involves models because they let us simplify what would otherwise be extremely detailed, complicated concepts by focusing only on essential elements. of… something. First, for anyone new to the concept of modeling (in this way), let’s talk about what it means to build a model.

A model is a simplified representation of something. It may simply be a mental thing. For example, we might have a mental model for humans: four-legged mammals that walk on two legs. If we see a four-legged mammal that’s not walking on two legs, then we decide that it’s not a human. (Yes, exceptions exist… and that’s one of the challenges of ML… which we’ll get to in the next post.)

In that model, there are a bunch of details that we ignore. Some of them are irrelevant: we have our own versions of claws (fingernails), so, other than their detailed shape, simply having those doesn’t help distinguish us from many other mammals. On the other hand, we have little body hair compared to most mammals. That could be a distinguishing factor, but maybe that’s one “rule” more than we need; maybe the walking-upright thing is enough for most purposes.

And that’s the thing: machine learning is never perfect; it’s simply good enough for whatever the task at hand is.

Training

The process by which a machine learns is, naturally enough, referred to as trainingWith machine learning, a model must be trained in order to perform its function (which is called inference).. Two broad approaches are used: so-called supervised learningA machine learning approach where a machine is trained by giving it examples and, during the training, telling the machine what the answers are. A large number of varied examples provide the best results. and unsupervised learningA machine learning approach where a machine is trained by learning as it goes. It doesn't get trained by examples ahead of time, and isn't explicitly taught the "right answers." There must be some other way of reinforcing the right and wrong answers as it learns..

Supervised learning means that we go through a specific training period where we use specific examples to train a systemThis is a very generic term for any collection of components that, all together, can do something. Systems can be built from subsystems. Examples are your cell phone; your computer; the radio in your car; anything that seems like a "whole.". Any IoT devices are most likely to undergo supervised training. For instance, if we were training a camera to identify a human – like our own mental model – then we might show lots of pictures of humans and other things, and we’d tell the system which are humans and which aren’t. Then we’d test the results using a bunch of pictures that weren’t part of the training to see how many it gets right. Poor score? Then look at what it got wrong and do more training using similar examples.

Quality Training

The quality of training has everything to do with how many examples are used and how varied they are. Let’s say we show 100 pictures of different people standing and 100 pictures of birds and fish. Well, the machine is likely to decide that humans are anyone that has four appendages, two of which aren’t wings. In other words, its model is too simplistic to work in many situations. That’s because we didn’t include enough variety in the training samples.

If, on the other hand, we included pictures of other mammals, now the machine is going to have to come up with other distinguishing features – like our walking upright. But our mental model comes from us seeing humans in lots of behaviors, so we might know that humans can walk on two legs – even if we see someone that isn’t walking right now – perhaps the person is sitting. We have years and decades of training that lets us recognize other human traits.

With a machine, it learns only what its presented with. The model it builds is mathematical; change one of the pictures in the training set, and the model might come out differently. Also, more than one model might work correctly: change the order of the images in the training session, and a slightly different model might result – one that would work just as well as some other model trained on those same pictures in a different order.

Next we’ll look at how we can apply training to the IoT.