Overview of Supervised Learning

In practical terms, AI is a set of algorithms that allow engineers to solve problems with a high degree of complexity. The reason to use AI is because it allows the solution to very complex, though specific problems.

Supervised Learning

AI offers the solution to a variety of different problems: for each defined category of problems we use what we call a paradigm. AI offers a solution to three defined paradigms: Supervised Learning, Unsupervised Learning, and Reinforcement Learning. Given their complexity, today I will focus on describing which are all the main Supervised Learning tools provided by the AI to solve this category of problems.

Supervised learning works when the data can be divided into Features (the data that we use as a predictor) and Labels (the part of the data we wish to predict). We start with a dataset that has both, and after the AI figures out the rules that binds them together, it can make an estimation on a new dataset.

However, there are several different configurations of our problem. Each one of these configurations has a specialized AI to solve that problem.

Let us begin by a clear representation of our dataset:

First of all, let us look at all the possible configurations to structure our Labels:

  • Single Label

A single column of labels (survivors of the Titanic)

  • Multi-Label

Multiple columns of labels (students IQ and grades)

The content of the labels for Classification:

  • Binary Classification

Labels can only have 2 values (cat or dog)

  • Multi-Class Classification

Labels have 3 or more values (cat, god, fox…)

And the content of the labels for Regression:

  • Regression

Multiple inputs, single output

  • Multi-Output Regression

Multiple inputs, multiple outputs

Combined, we obtained the following scheme: it summarizes all the possible ways we can classify a problem we wish to solve using AI (using Machine Learning algorithms or Multilayer Perceptron neural networks). 

Practical Examples

  • Iris classification

The dataset requires a classification approach, it has one single label, 4 kind of iris:

Classification, Single-Label, Multi-Class Classification

  • Titanic survivor prediction

The dataset requires a classification approach, it has one single label (if a person survives or not), 2 kinds of choices (alive or dead):

Classification, Single-Label, Binary Classification

  • Predict student grades with their IQ

The dataset requires a regression approach, it has two labels (grade of students and IQ), we require one output for both predictions:

Regression, Multi-Label, Regression

Related Posts

Leave a Reply