How Machine Learning Works and What are its Types. “Hey Siri, play Lady Gaga,” we say to our phone, and it tells us that ok, it’s going to put Lady Gaga on Apple Music. The phone is not only executing a command but it analyzes and stores it to identify what kind of music you like. In this way, if the next time you ask Siri (Apple’s voice assistant, but it works similarly with Alexa or Google) to play music for you, without specifying which one, the assistant will select music in the same style as Lady Gaga.
We could put another more complicated example. Let’s imagine a big city, where traffic problems are common at peak hours. The information sent by mobile phones is used to determine the points of greatest congestion, data that is analyzed together with the images sent by cameras located on streets, highways, and ring roads. All data processing is carried out by a software system to which smart traffic lights are also connected.
The result is that the system, based on the analysis of the data it receives, executes traffic light times, managing preferences at a crossroads, in one direction of travel over another, pedestrians over cars, etc. But what is really significant about this example is not that the traffic system adjusts the traffic lights when there are traffic jams,
When we use Netflix, the algorithm is constantly learning from our choices, to show us the content that we like the most. And the same thing happens with search engines like Google or channels like YouTube, which constantly learn from our searches, what option we select, how long we have watched the content, etc.
All of this is done with Machine Learning technology, a branch of Artificial Intelligence (AI) and computer science that uses data and algorithms to mimic the way people learn. And it does it automatically, through experience, without having been expressly programmed to do so. Machine Learning is directly related to other technologies, such as big data or business intelligence.
How Machine Learning Works
During its development, the software is fed by specific data sets, from which it learns. In this initial training phase, the software is fed with instructions, examples, and the experience it accumulates, which will allow the algorithm to find patterns and make its own decisions in the future and in new circumstances, learning more and more and without human intervention.
Types Of Machine Learning
Automatic learning of computer systems is carried out mainly in three different ways, depending on the degree of intervention of people in its programming and configuration.
Supervised Machine Learning
Anti-spam filters use this model, in which we constantly tell the software what words, expressions, or characteristics of the message indicate that it may be spam. As we enter this tagged data, the filter becomes more precise. Emails identified as spam are sent to a different folder than the inbox. An example: if the email contains the word “sex”, 3 points are assigned; if it contains the word “Viagra”, 2 points are assigned; if the word “promotion” appears in the subject of the message, 2 points are assigned. If the score exceeds 6 points, the message is marked as spam and sent to the spam folder instead of the inbox. We previously introduced the labels «sex», «Viagra» and «promotion», which is why we speak of supervised learning.
Unsupervised Machine Learning
In this case, the algorithms analyze and group unlabeled (unidentified) data sets, discovering hidden patterns without the need for human intervention. Examples of this type would be image recognition or data analysis of thousands or hundreds of thousands of customers, from which the system can segment according to the patterns it identifies.
For example, a system can classify images based on the objects that appear in them without having been given specific instructions to do so, without having introduced labels in the system.
In this modality, during the training phase of an unsupervised system, it can be assisted with a smaller set of labeled data to aid in classification and pattern extraction. And on the other hand, the insufficiency of labeled data to train a supervised learning algorithm can be compensated.
As you can see, Machine Learning is a technology that is already used in many solutions that we use on a daily basis. Now the challenge is to make it available to professionals and SMEs, showing the benefits it brings to businesses and providing training so that they know how to use these tools.