How machine learning learns from examples — Technology, 11–13 years
Some computer systems improve by finding patterns in examples rather than following one list of instructions. Their results depend strongly on the examples they receive.
Learning patterns from examples
In machine learning, people give a computer many examples and a way to measure whether its guesses are useful. The computer adjusts its internal settings to notice patterns, such as shapes in pictures or words in messages. It is not learning like a person; it is calculating from data.
Why use examples
People use machine learning when writing every rule would be too difficult or the world changes too often. Recognising speech, sorting photographs and spotting unusual payments involve countless variations. Examples let the computer find a useful pattern, but they do not guarantee a correct answer every time.
Teaching a picture sorter
A system must sort apples and oranges. First, it sees 100 labelled pictures: 60 apples and 40 oranges. It notices colour, roundness and surface patterns, then guesses the label of a new picture. If it gets 18 of 20 test pictures right, its accuracy is 90 percent, but the test pictures must be separate.
The cleverness trap
A common mistake is assuming that a confident computer answer must show understanding. A model can be good at matching patterns while missing the meaning, and it can learn a misleading feature from the examples. This is reasonable because fluent answers look thoughtful, but confidence can be just a calculated score.
Where it appears
Machine learning appears in photo searches, speech tools, recommendations and systems that flag possible fraud. It can save time when patterns are large and repetitive. When a decision affects a person, such as a medical or school decision, people should question the data, errors and responsibility behind the result.
Keep exploring
Other languages
Loading MyLeoNes™…