MMOOZZGG.com

Glossary

What is machine learning?

In short

Machine learning (ML) is a field of AI in which a program finds rules in data by itself instead of getting them ready-made from a developer.

AI basics · machine learning

MMOOZZGG.com editorsChecked: 1 min read

For a child

How did you learn that a cactus is prickly? Nobody wrote you a rule: you touched one once and remembered. A computer can also learn from experience. You show it many examples with answers: this is an apple, this is a pear. Then it sees a new fruit and guesses by itself what it is. The more good examples, the fewer mistakes it makes.

In plain words

Usually a developer writes rules: if an email contains the word prize, send it to spam. But scammers change the words, and the rules soon go out of date. Machine learning works differently: you collect thousands of emails marked spam and not spam and let the algorithm find the differences by itself. The result is a model that judges new emails too. Movie recommendations, credit risk scoring at a bank and traffic forecasts in a navigation app work the same way.

For business

This approach helps where a company has built up a history: sales by day, requests, calls, payments. On it you build a demand forecast, an estimate of the chance a customer will leave, a search for suspicious transactions, and next-step hints for a manager. Costs depend less on the algorithm than on the data: it must be collected, cleaned and labeled, and this often takes most of the budget. Risks: the algorithm learns from the past and makes mistakes when the market changes, and it repeats any bias in the history. Before launch, accuracy is checked on a recent period the system has not seen, and it is monitored afterward. Typical scenarios are covered in the section on predictive analytics.

For a specialist

Machine learning, as defined by T. Mitchell (textbook Machine Learning, 1997), studies algorithms that improve automatically with experience. The main modes: supervised learning on labeled pairs of object and answer, which includes classification and regression; unsupervised learning for clustering and dimensionality reduction; reinforcement learning, where an agent gets a reward for actions. Classic methods: linear models, decision trees and their ensembles (random forest, gradient boosting), support vector machines (SVM), k-nearest neighbors. Deep learning uses multi-layer neural networks and extracts features by itself, while classic methods usually need features prepared by hand (feature engineering). Quality is measured on a held-out sample (test set) and with cross-validation; the main danger here is overfitting.

Example

A pharmacy chain loads its sales for past years, and the model forecasts how much of each drug every pharmacy will need next week. Buyers write off expired goods less often.

Related terms

Questions and answers

What is machine learning in plain words? It is a way to build programs that learn from examples. Instead of writing every rule, you give them a history with correct answers, and they find the patterns themselves.
How is machine learning different from neural networks? Neural networks are one of the methods inside this field. Many practical tasks, such as a sales forecast from a table, are solved without them: with decision trees or linear models.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us