Where Do We Encounter Supervised Learning?

• Labeled examples
• A model
• A prediction

What you will learn

• Explain what supervised learning is and what a label is.
• Name places where we meet supervised learning in daily life.
• Tell classification from regression and say what a trained model does with new data.

Before you start

It helps to have read the first article about artificial intelligence, but you only need to know what a computer program is.

What is supervised learning?

Supervised learning is machine learning that uses labeled data. Models are trained on examples with answers so they can learn to predict labels; accuracy on new data still needs to be checked.

Think of flashcards with a picture on the front and the answer on the back. The answer is the label.

MIT Sloan gives an example. An algorithm is trained with pictures of dogs and other things, all labeled by humans. The machine then learns ways to identify pictures of dogs on its own.

Each example has two parts. Features are the values that a supervised model uses to predict the label. The label is the answer, or the value we want the model to predict.

• Features
• The label

Where do we meet supervised learning in daily life?

Some examples of supervised learning are identifying spam and predicting precipitation, two tasks Google names as examples.

IBM adds that supervised machine learning helps organizations solve real-world problems at scale, such as classifying spam or predicting stock prices.

Pictures are another place. IBM describes an image classification model created to recognize images of vehicles and determine which type of vehicle they are.

Weather shows how features and a label fit together. In a weather model that predicts rainfall, the features could be latitude, longitude, temperature and humidity. The label would be rainfall amount.

• Filtering spam
• Recognizing images
• Predicting rainfall

What kinds of tasks does it solve?

Supervised learning tasks can be broadly divided into classification and regression problems.

Classification uses an algorithm to sort data into categories. Sorting messages into spam and other mail is one case. Recognizing a type of vehicle in a picture is another.

In regression problems, the output is a continuous value. A prediction of an amount of rain is one example, because the answer is a quantity and not a category.

• Classification: sorting into categories
• Regression: predicting a continuous value

What does a trained model do with new data?

The model is used to predict outputs for data it has not seen before. IBM states the goal of learning as correct predictions on new real-world data; that is a goal to test, not a guarantee for every prediction.

That is why a spam filter is useful. The training examples are already labeled, while the new messages that arrive each day are not.

• Labeled training data
• A trained model
• New, unseen data
• A prediction

Real-world example

Try this on paper with an adult. Write imaginary messages, such as a meeting invitation and an unwanted advertisement, then label each one "ordinary email" or "spam". The labels are the answers for these examples; there is no need to open real messages or click links.

Common misunderstandings

Supervised learning is a rare special case. In its overview, MIT Sloan described supervised machine learning as the most commonly used type at the time the article was published.

A model only repeats its training examples. The goal is a model that can predict correct outputs on new real-world data.

A label is always a name or a category. In regression problems the output is a continuous value.

Frequently asked questions

What is an example of supervised learning?

Identifying spam is one. Google also names predicting precipitation, and IBM names classifying spam or predicting stock prices.

What is a label in machine learning?

A label is the answer, or the value we want the model to predict. In a rainfall model, the label would be rainfall amount.

Does supervised learning need labeled data?

Yes. In conventional supervised learning, training examples include target labels or values. Self-supervised learning is a related but separately named approach that derives its own training signals from data.

Summary

• Supervised learning trains a model with labeled data so that it learns to predict labels, and features are the values it uses to do so.
• We meet it in identifying spam, recognizing images of vehicles and predicting rainfall.
• Tasks divide into classification, which sorts data into categories, and regression, which outputs a continuous value. The goal is correct predictions on new data, but results still need to be checked.