Supervised Learning

Supervised Learning

Supervised learning is the most common form of machine learning, and the one behind most of the AI you encounter day-to-day.

The idea is simple: you teach the model by showing it labelled examples — input/output pairs — and it learns to predict the output for new inputs it hasn't seen.

The word "supervised" comes from the fact that a human has labelled the training data. The model is learning under supervision.

How It Works

Training data consists of pairs:

• Input: a photo → Output: "cat" or "not cat"
• Input: an email → Output: "spam" or "not spam"
• Input: a patient's medical readings → Output: "diabetes risk: high / low"
• Input: a house's features → Output: predicted sale price

The model learns the relationship between inputs and outputs. Given a new input it's never seen, it applies what it learned to predict the output.

The quality of the labels matters enormously. Garbage labels → garbage model. This is why data labelling (often done by humans at scale) is a huge industry.

Classification vs Regression

Supervised learning problems come in two main flavours:

Classification — predict a category

• Is this email spam or not spam?
• What digit is in this image? (0–9)
• Does this tumour look malignant or benign?

Regression — predict a number

• What will house prices be next quarter?
• How many minutes will this delivery take?
• What's the expected lifetime value of this customer?

The same learning approach applies to both — the difference is in what kind of output you're predicting.

Where You See It

Supervised learning is behind:

• Spam filters: trained on millions of labelled emails
• Image classifiers: trained on ImageNet (1.2 million labelled photos)
• Medical diagnosis tools: trained on labelled scans and patient outcomes
• Credit scoring: trained on historical loan data with known outcomes
• Speech recognition: trained on audio paired with transcripts
• Large language models: trained on text with the next word as the label (a form of self-supervised learning, which we'll cover later)

The vast majority of practical ML applications use some form of supervised learning.