I thought we should tackle one of the first questions asked in machine learning:

“What is the difference between supervised and unsupervised machine learning?”

In the machine learning field, there are two subcategories of machine learning called “supervised” and “unsupervised” learning. No, this doesn’t mean you have to watch supervised models run and you can read a book while an unsupervised model runs. Let’s take a look at what these terms mean and how we might use each of them in data science.

We are going to explain the difference between supervised and unsupervised machine learning using a simple shape sorting toy you probably played with as a child, the shape sorting cube.

Supervised Learning

Lets say you wanted your child to learn how to sort the shapes. A supervised approach may be laying out all of the shapes, telling the child what each shape is (i.e. “This green one with 4 straight sides is a square”).

After having each of the shapes “labeled” (and probably repeating this a number of times), the child learns that squares have certain features (4 sides that are straight) and are able to find the corresponding shape in the box (the hole with 4 sides that are straight).

This is similar to supervised machine learning, where we give the machine labeled data, meaning we give it the answers. By repeating this a number of times, the machine learns the features that result in a labeled outcome (in this scenario, the features would be 4 sides and straight sides, or 90 degree angles at each point). Supervised machine learning is used in many classification problems (this is a “square” vs. a “circle”) and in many regression problems (we predict the weather will be 40.3 degrees today). Both of these are problems where we train the model using a data-set containing labels of the outcome (shape for our classification problem and temperature in degrees for our regression problem).

Unsupervised Learning

Another approach to learning the cube sorting task is to learn by observation and pattern recognition. In this type of learning, you wouldn’t tell a child which shapes were which. You would let them explore the differences on their own, coming up with their own idea of the features. In this scenario, perhaps the child clusters them by groups: all of the shapes with 4 sides, for example. They may break this group down even further to be 4 straight sides (square) vs. 2 straight sides and 2 slanted sides (trapezoid). In machine learning we call this “clustering”, and there are many different methods for doing this!

#ai #machine-learning #supervised-learning #data-analysis #data-science