Data: what computers learn from

AI learns from data. So what exactly is data? Data is information that a computer can store and use. Numbers, words, photos, sounds and videos can all be data. A I learns from data, a bit like the way you learn from experience.

Kinds of data

Numbers, like prices, temperatures and exam scores. Text, like messages, books and websites. Images, like photos and medical scans. And sound, like voices and music.

New word: Label

To learn, many A I systems need labelled data. A label is a tag that says what something is. This photo is labelled ripe mango, and that one, unripe mango.

What makes good training data

Lots of examples. More good examples usually means better learning. Variety. Pictures taken in different light, from different angles, in different places. Correct labels. If the labels are wrong, the A I learns the wrong lessons. And data that was collected fairly, with people's permission. Every time you use an app, it may collect data, like what you watch, search for, or tap. Companies use that data to improve their A I, and to show you adverts. That's one reason your privacy settings matter.

Quick check

Quick check. Which of these is data? All of them. Photos, voice notes and messages are all data. Options: All of these: a photo, a voice note and a text message; Only numbers; Only books; Nothing on a phone.

Recap

Data is information that computers can use. It can be numbers, text, images or sound. Labels say what each piece of data shows. And good training data is plentiful, varied and correctly labelled.

I've read this lesson