unit 0.4 - Types of learning

There are different types of learning in machine-learning, depending on how a desired relationship is acquired.

Supervised

\(y_d = f(x)\) with \(y_d\) as desired output and \(x\) as input.

Samples in the dataset are {{x1,y1}, {x2, y2}, {x3, y3}} – both \(x\) and \(y\) are provided as examples!

Examples:

  • a picture of a cat and a category number (1 = cat)

image0

  • from a list of homework grades (a vector \(x\)) to final class grade \(y\)

  • generating an image from a text caption (Dall-E)

image1

Unsupervised

There is only \(x\) here! Can we split \(x\) in groups?

image2

Examples:

  • a zoo of animals, all of different kinds - group by kind

  • a set of points, group by closeness (k-means clustering)

image3

  • a set of people, group by age, race, sex, …

Self-supervised

There is only \(x\) here! What is the desired output \(y_d\)? It is \(x\) itself!

Examples:

  • A portion of an image from the rest of the image (in-painting)

image4

  • Removing noise from an image (de-noising)

image5

  • predicting the next words from a previous set of words (language models like: chat-GPT, Gemini, Llama)

image6

Reinforcement learning

Learning to improve a task based on a performance metric. Self-improvement by comparing your results and trying to optimize gains.

image7

Examples:

  • a program that learns to play chess, go, tic-tac-toe, etc. from playing the game

  • an athlete that wants to improve its moves to achieve better results

  • a robot that learns to pick-and-place objects by trying not to drop them, or improving speed

Imitation learning

Copy a sequence of items from one instance to another. Learning to do something by observing someone else.

image8

Examples:

  • learning to draw a figure by observing someone

  • learning to play music from a youtube video

  • learning a recipe from a video

  • learning to code machine-learning from video-lectures

  • learning to sing after someone

image9

Summary

These are the most typical learning techniques and data samples in the respective datasets. The examples of applications are innumerable and we only mentioned a few here for brevity.

References

Just a few links here but there are many more. Search online for β€œtypes of learning”.

1-types of learning

2-types of learning