3 (or 3.5) Types of Machine Learning – Supervised, Unsupervised and Reinforcement Learning

3 (or 3.5) Types of Machine Learning – Supervised, Unsupervised and Reinforcement Learning

Well. We all have to learn from time to time. In fact, you are about to learn something new. 


As kids we learn that we need to say “please” and “thank you” – our parents made sure to remind us of that. As students, we had to study for our subjects and as adults, we probably have to learn new skills to stay relevant in our job.  


We may all have different methods to learn.  


So does AI.  


Not the only ones, but the most common learning methods are: Supervised Learning, Unsupervised Learning and Reinforcement Learning.  


Let’s investigate those a little bit.  


In supervised learning, we provide the model with labeled data. This means that with each dataset a solution or answer is provided.  


So, if we give the model a bunch of images of fruits and we tell the machine which kind of fruit the images show. For example, a banana, an orange, or a strawberry.  

When we provide the model with a new image it will be able to predict the label correctly because it compares it to the input data, the labeled images of fruit.  


Easy, right? 


Supervised learning is useful for classification problems, like the fruit example. But it can also be used for so called “regression problems”.  This is when data is not discrete, so does not have whole number, but continuous data. For example, predicting the rental prices of apartments in certain countries, cities or districts.  


Okay, great let’s continue with Unsupervised Learning.  


The biggest difference between those two learning methods is that unsupervised learning model receives:  

Unlabeled data 


So, the model would not be able to know that the image of the banana shows an actual banana, because it was not labelled this way.  


But how does the model learn then and guess the output? 


By finding patterns, similarities or groupings within the data.  


It groups bananas correctly, because it recognizes that they are yellow and curved.  

It groups oranges correctly, because it recognizes that they are orange and round. 

And the same goes for strawberries because they are red and have a green leaf. 


This is how the model, can distinguish the fruits from each other. 


There is also a hybrid version called semi-supervised learning, where a a small amount of labeled data and a large amount of unlabeled data is given to the model.  


Alright. If you paid attention, you know that we covered two out of the three learning methods we want to cover in this article.  


Now we will look into reinforcement learning, which has nothing to do with the previous two learning methods.  


We find reinforcement learning e.g. in robotic vacuum cleaners or autonomous driving. 


It learns based on interacting with its environment and by trial-and-error.   


Let’s take the robotic vacuum cleaner as an example.  


Their goal: clean the floor of your apartment. 


When you set it up, it does not know the correct paths to take yet to achieve its goal. So, it moves around and learns by trial and error. If it bumps into something, and the sensors are triggered, it rotates and moves forward until it finds a clear path. Over time, it knows how to navigate through your apartment without bumping into obstacles.  




Let’s summarize this. 


Supervised Learning = through labeled input and output data, models can learn to predict the correct output when given a new input  

Unsupervised Learning = machine organize unlabeled input data by finding patterns and similarities 

Reinforcement Learning = machines interact with its surrounding, execute actions and learn by trial-and- error  


And of course, let’s not forget about semi-supervised learning, where the model uses a small amount of labeled and a big amount of unlabeled data for training. 


Well done.  


Everyone who learns also needs to take a break. Enjoy.

Back to blog

Leave a comment

Please note, comments need to be approved before they are published.