Machine Learning for Dummies

Machine Learning is a way to explain a computer what to do. You can’t just direct summon a computer do perform a task. That is why you give instructions on how to recognise relations and how features are recognized. By ‘teaching’ a computer, we can use these devices to help us with tasks that people are good at, but computers traditionally were not.

For people, it is easy to recognize objects or people, however for a computer that is much more difficult. In fact, a computer is a stupid device. Everything you can do with it, needs to be told to the computer first. An app tells the computer what should appear on the screen or what should happen when you press a certain button. Everything a computer does is conceptualized and tested by programmers in advance. And because a computer cannot do anything on its own, that is pretty difficult.

If a programmer makes a small mistake, like 1 + 1 = 3, the computer will consider that answer to be the truth and it will always keep giving the ‘wrong’ answer. Yet, it is possible to have a computer to do complicated things; play games, write letters, take and edit photos, you name it. Before a person can explain something to a computer very precisely, we must first completely understand it ourselves.

A recognizable example; Everyone understands what a chair is, but it is not so easy to explain that to a computer. Most characteristics of a chair are also attributable to other objects. For example, “it has 4 legs” can also be a table but an office chair has only 1 leg with 5 wheels, or “you can sit on it” that can also refer to a bean bag or sofa. And this is only one object, there are countless objects, subjects and connections that have similar and also different characteristics. Hence, it is almost impossible to tell a computer what a chair is. The solution is to show a computer numerous examples of a chair and thus to ‘teach’ it to recognize a chair.

By now, we sufficiently understand how to teach something to a computer, or even better, explain how it can teach itself. So instead of telling the computer what a chair looks like, we explain to the computer how it can learn from examples. We show examples of chairs, and of things that may look like a chair but are not a chair, until the computer recognizes almost all chairs.

In addition to recognizing or classifying objects, there are other tasks that are easy for human beings but difficult to explain to a computer:

• grouping (clustering): We want the computer to sort things so that it can tell us which things belong together and why. Without indicating in advance which criteria we want to use.

• predicting (regression): The computer can make predictions for the future, based on examples. For example, it can tell you what the price of a barbecue or inflatable pool will be if next week’s weather forecast is looking good. Without us telling the computer about the correlating calculation. The computer has ‘learned’ from the examples and can thus determine whether and to what extent the weather influences the price.

The machine has learned to recognize objects, make predictions or to group. The factually stupid device has received enough information to teach itself these skills. The calculation speed of the computer and the ‘thinking’ power is many times greater than that of a human being and thus the chances and possibilities of a computer are proportionally greater.

This article was originally written for and published at SMT