Machine learning belongs to the study field of Artificial Intelligence. Focuses on creating a program that is capable of gradual improvement in a specific activity based on data. The more data can be provided, the better the program can learn. The whole benefit of machine learning is that it can learn without being taught by anyone. What can we imagine when we say "teach a program"?

Classical programming works by typing a set of rules that the program should follow on the "if-then" principle. If something comes to the entry and the program does not recognize it (or what the programmer forgets about), it cannot move further.

Machine learning, however, is different. It passes through all the data, and finds a similarity or a template that makes it easy to memorize and learn. This is a huge advantage over classical programming. When there are many variables, one is not able to think of all of them and set the program taking everything into consideration. Also, the more variables we have, the more robust the program becomes and more exacting for a hardware demands or maintenance.

What are the methods of Machine learning?

There are many of them but I will only describe a selected for a better understanding of this Machine Learning trend.

1. Supervised vs. Unsupervised learning
The difference between them can be beautifully explained by the example. Imagine having a younger brother and wanting to teach him letters. You will show him the letter D and tell him that he will recognize the "D" by its specific "belly" shape. If he then points to another letter, such as "P", and tries to apply the rule he learned (if the letter has a "belly" then it’s D), then the teacher has to come in and correct him.

But what does that mean in machine learning? In supervised learning the program is given a set of data where we know the inputs and outputs. By outputs we mean historical data (what, why and what was the result) and hence the program based on the principle of similarity can predict of what will happen. It can be compared to a human expert who has been working for many years in a specific field, and who, based on his experience, can tell what's going to happen.

Another question that popped up immediately after this was: Is not then the teaching with the teacher always better? Not so completely. Sometimes we do not have the output data; sometimes we can have too broad and abstract problem, where it is impossible to say what the result is.

2. Reinforcement learning
This is the learning I personally work with as part of my bachelor thesis and that is why I am close to it. It works by way of reinforcement for a good decision and a way of punishment for a wrong one. Simply put, we have several different options that we can do (as many as 100 options).

Reinforcement learning make a lot of decisions and go through all the options (so called iterations/ epochs) and thanks to a lot of experiments it will learn or find the best possible combination and sequence of steps based on how many times it was reinforced and punished.

What are the options of use of machine learning?

Those are countless. I think the only thing that can possibly limit us is our imagination. When we recognize the strengths and weaknesses of machine learning, finding the right focus is then just a piece of cake. I will only mention a few that are actually being used or are interesting from an information or financial point of view.

Personalization of a person on the Internet
Collecting data about the customers, which products they are viewing, which articles interest them - all of this can be used to deduce what they want to solve, what problem they have and based on the data from other users with similar problems the program can advise and suggest products which could be of help.

Google or Facebook work on the same principle. They, on the basis of the so-called Cookies, collect the data about the person and give them the most relevant results.

Online customer line
So far, people have always done it. Many companies offer phone contact through customer hotline service. High utilisation of people, frequent repetition of the same questions resulted in the use of machine learning along with so-called chatbots (a program that replies on human questions in real time). You might think that no computer could advise you correctly, or that as a person you could know the difference immediately. But the well-known Touring Test has already convinced us several times that artificial intelligence may be indistinguishable from humans.

These are just some of the many uses of machine learning and the principle of its functioning.

Tomáš Lichanec

Technical university of Košice
Faculty of electrical Engineering and informatics, Intelligent systems Machine learning