It often feels like there is a lot of jargon floating around when it comes to technology. There are so many terms for so many unique things, but then again it can sometimes feel like there are so many terms for exactly the same things.
It’s fair enough, some terms do mean similar things and sometimes it can get confusing when one thing is an extension of another. To try and clear up some confusion, here’s a list of the terms we come across most often, with neat definitions and examples.
And then, to see if you really understood it all (or that we really explained it right), test your smarts in the ultimate tech quiz.
The word used for anything where the computer is emulating intelligence. It is often used in very narrow fields, such as a computer that tries to navigate a user from point A to B through a city. Like Dijkstra’s shortest route algorithm for GPS navigation, it can choose the most attractive option but cannot learn to adapt to user input.
A subset of AI, machine learning uses thousands of data points to change the strategy slightly and alter future behaviour depending on how users react. For example choosing routes that are travelled more often by a user or one that avoids traffic. We see this in more advanced map applications (think Google Maps).
A type of Machine Learning, that simulates the way the human brain works. With a LOT of small ‘neurons’ that each have a very specific task. An example of this could be an algorithm that tries to identify whether there is a hotdog on a picture or not. Where one specific neuron might look for shades of red in a specific format. It’s important to note that it’s not humans that define what each neuron looks for. It’s entirely the computer that finds the best ‘settings’ for each neuron.
An extension of machine learning, where there are many layers used in the decision process. Using millions of data points, deep learning can determine more subtle relationships between data points, like the change of a brush stroke between two paintings.
Just to make it even more fun, when governing the types of machine learning (and as an extension deep learning), there are different rules that are used, depending on the aim of the algorithm.
These fall into two categories: supervised and unsupervised learning. And of course there are subcategories that branch off from them, but don’t worry we have you covered there too.
The data given to the computer has both some input and a desired output. So if we are looking for hotdogs, we would have the picture (input) and whether there is a hotdog on it (output). The machine then teachers itself based on these pictures, so it can give an answer on a picture it hasn’t seen before. This is the kind of thing that is used when you filter out spam emails in your inbox.
Less guidelines are given to the computer so that it is missing some data points. This requires it to make assumptions in lieu of complete data. So, if a machine is told to observe objects moving at 20 km/h, it will not be able to tag objects that move at other speeds but it will still notice them.
Active learning is a part of semi-supervised learning where only some data-points are labeled and other labels are added as the programme learns. Think about when you are asked to check a box to prove you are not a robot, when you select images that match the question (e.g. all images containing cars) you are helping to train the programme.
Here a computer is learning against itself, improving functions through a series of experiences. Think for example of the game Pong, if you had four versions of a computer playing against itself (tournament style) the two winners would then be pitted against each other, and the final winner will, in theory be the best player. This can be repeated numerous times to eliminate weak spots in the way the computer plays the game, eventually making it an unbeatable player.
Here no labels are given to the data so the computer has to try and derive patterns alone. In this way hidden patterns may be discovered that may not be intuitive at first. This works really well when forming recommendations for webshops, for example it may seem nonsensical to recommend perfume and footballs together but by analysing data with unsupervised learning, the connection suddenly becomes apparent - mothers and sons frequently buy together in the webshop.
So do you think you’re a tech wizard now? Take the test to find out!