Chapter 8 Machine learning types
“If intelligence was a cake, unsupervised learning would be the cake, supervised learning would be the icing, and reinforcement learning would be the cherry” Yann LeCun
Three main ML types are defined based on the way the work. There are also algorithms which are a mixture between supervised and unsupervised learning which are called self-supervised algorithms. For example in natural language processing (NLP) text can be used to train models by omitting one word and train the algorithm to predict the omitted word.
There are three major types of ML
- Supervised learning
- needs labeled data
- Unsupervised learning
- finds groups with similarities
- Reinforcement learning
- finds a policy to achieve a goal
Examples for ML types
In the following interactive graph application areas for ML types are given to create an understanding on what can be done with the different types. The applications range from dimensionality reduction to targeted marketing.
- click at one item the neighboring item will be highlighted.
- use the pull down menu to find a list of all entries
- to zoom use mouse or touch pad
- move item by click and drag
- control elements at bottom of graph can be used to control graph as well
8.0.1 Machine learning types by data type
It is also possible to look at the ML types from the perspective of data types. There are a variety of data types for in- and output. The output type does not have to be the same as the input type. For example image as input can have text or speech as output.
8.0.2 Machine learning types by data content
Another perspective of the ML types is the view on the data content. It might be advantageous to create a heatmap of mouse movements during a credit application and use that heatmap as one indicator to decide about the application. Another example is to convert sound to short time frequency plots to determine what the sound represents.
In the following chapters each type will be introduced and examples are given.