What is machine learning?

The definition of machine learning, its three types and the idea of generalization.

20 minutesIntroduction to AI without programming

What you will learn

  • Explaining the difference between supervised and unsupervised learning
  • Understanding why training data is separated from test data

Machine learning is a branch of AI in which the algorithm learns a pattern from past examples in order to predict new ones. Instead of writing the rule “if the petal length is less than 2.5 cm, the flower is Setosa”, we give the algorithm a table of measured flowers and their species, and it infers the rule by itself.

Its main types: supervised learning (Supervised: we have correct answers, for classification or regression), unsupervised learning (Unsupervised: no answers, for clustering and dimensionality reduction), and reinforcement learning (Reinforcement: an agent learns through reward and punishment).

The most important concept: the model is evaluated on data it did not see during training, otherwise we may fool ourselves with an illusory accuracy.

Implementation steps with the tool

  1. In Orange: File (iris), then Tree, then Test and Score, and note that the accuracy is computed on instances the model was not trained on.

Quick review question

If we have a customer table without any target column, which type of learning do we use?

Tools

Related algorithms

Try it in the lab