Introduction to AI 2

by Albert Major

This slide explains the difference between **supervised** and **unsupervised learning**, two major types of machine learning covered before introducing **reinforcement learning**. In **supervised learning**, the model is trained on **input–output pairs** (e.g., images with labels like “cat” or “dog”), learning to map inputs to correct targets using methods like **backpropagation**. In **unsupervised learning**, there are only **unlabelled inputs**, and the model tries to find **patterns** or **clusters** (e.g., grouping cats and dogs separately without knowing what they are). This sets the stage for reinforcement learning, which is fundamentally different because it doesn't rely on datasets with labels at all.

1. Machine Learning Types

Supervised Learning

• You are given pairs of inputs and outputs (called "labels").
• Example: Pictures of animals + their correct labels ("cat" or "dog").
• The goal: Train a model that can predict the correct output for new inputs.
• How it works:
  • Start with random predictions.
  • Compare model output with the correct label.
  • Use a method called backpropagation and gradient descent to adjust the model so it gets better over time.

Unsupervised Learning

• You only have inputs, no labels.
• Example: Download millions of random images from the Internet.
• You do not know what is in each picture.
• Goal: Find patterns, such as grouping similar images together into clusters.
• You might discover that cat pictures are naturally more similar to each other than to dog pictures, even without labels.

Self-Supervised Learning

• A clever way to create labels from the data itself.
• Example: Turn a color photo into black and white.
• Then train a model to predict the original color photo from the black and white version.