[From the last episode: MachinesIn our context, a machine is anything that isn't human (or living). That includes electronic equipment like computers and phones. can be trained to do tasks that have traditionally been hard for machines but easy for humans.]

TrainingWith machine learning, a model must be trained in order to perform its function (which is called inference). doesn’t happen on an actual IoTThe Internet of Things. A broad term covering many different applications where "things"  are interconnected through the internet. device – it takes way too much computing for that. So the training is part of the development of some algorithm when some new device is being designed. Once the training is complete, the 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. is built into the IoT device for shipping. That uses less circuitry than a training system would.

This is one area where updated your devices can help. You buy a device with some amount of training. But, even after you own it, the company that designed it may be refining its training – perhaps adding new examples that help in areas where they found the model wasn’t working. Those exceptions I mentioned in the last post could be an example of this: adding pictures of seated humans if the company learns that such images are confusing the original model.

So the 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. improve over time; this is why you might be able to get a softwareIn this context, "software" refers to functions in an IoT device that are implemented by running instructions through some kind of processor. It's distinct from "hardware," where functions are built into a silicon chip or some other component. update that includes a better model. If your device is having trouble getting some task right every time, then it’s good to install the updates. (It’s always good for improving securityRefers to whether or not IoT devices or data are protected from unauthorized viewers., but the device might actually work better after an update.)

Training for Automobiles and Security

Two areas where 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. can help are in self-driving – or even advanced human-driven – cars and security.

Let’s take, for instance, the case where your car has a feature that tells you when you’re drifting out of your lane. To us, as long as we’re paying attention, it’s obvious: we’ve got those lines telling us where we are on the road. But to a machine, it starts out knowing nothing of lines. So it has to see lots of video examples showing staying in the lane and going out of the lane for it to learn when to sound the alarm.

Security gives a couple possible flavors. For a security camera, we have the classic challenge that has given rise to the fogRefers to some way of doing "cloud"-style computing without having to use the cloud. A local server (or router or something else) can act like the cloud; because it's local, it's referred to informally as the fog., which we recently discussed. The camera will generate lots of video data that we need to analyze, and that analysis might be a lot of work. So, either we pay for sending all that data 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. where the heavy lifting can be done, or we add more computing power to the camera (or some local device its attached to), increasing its cost.

Depending on what you’re looking for, it might be possible to construct a simple-enough model that it won’t make the camera too expensive. So the training for that 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." would consist of images (or video) of things that are threats and things that aren’t.

Going Unsupervised

For other kinds of security, the other type of training – unsupervised – can also work. Let’s think about networkA collection of items like computers, printers, phones, and other electronic items that are connected together by switches and routers. A network allows the connected devices to talk to each other electronically. The internet is an example of an extremely large network. Your home network, if you have one, is an example of a small local network. security: what if someone is hackingThis can mean a couple things. A quick-and-dirty (but not elegant) trick to get something done is a hack. A computer security break-in is also a hack (because inelegant tricks are used to break in). It can be a noun or a verb ("he hacked my computer"). in and fishing around to find some tasty data. How do you train a system to know what’s allowed and what isn’t in that case? It’s really hard – practically impossible – to identify all the things that could be threats. HackersA misused, but common term for an unauthorized person trying to break into a device or network. Originally, in this context, "hackers" referred to the good guys (or "white hats"), while "crackers" were the bad guys (black hats). are specifically coming up with new ideas all the time.

With unsupervised training, you might say that the system learns as it goes. Which makes it mostly not viable for small, battery-powered items. But just to show how it’s different, one way to describe it might be that it learns what “normal” looks like. Then, if something happens that’s not normal, it can flag it.

For example, let’s say that it learns that it’s normal for five specific individuals to be on the network. If that’s all it’s looking for, and if you hire someone new, it’s going to flag that new person on the network as a problem – until you pat its worried little head and say, “No, it’s fine.”

Caveats, Caveats

I should emphasize here that ML is a very active area of development. Different companies may use different techniques, and there are different kinds of computing chipsAn electronic device made on a piece of silicon. These days, it could also involve a mechanical chip, but, to the outside world, everything looks electronic. The chip is usually in some kind of package; that package might contain multiple chips. "Integrated circuit," and "IC" mean the same thing, but refer only to electronic chips, not mechanical chips. that do better or worse, depending on the exact algorithm being used. The descriptions I’ve given have been very high-level; any details would likely change over the next five years as companies learn how to do this better.