ELI5: unsupervised learning

What is unsupervised learning?

Unsupervised learning is when a computer program learns to find patterns in information all by itself, without anyone telling it what to look for [1][2]. It's like giving a kid a big pile of toys and asking them to organize it however they think makes sense—the kid figures out that balls go together, blocks go together, and stuffed animals go together, all on their own [5].

Why does the computer need to figure it out?

Sometimes we have tons of data but nobody has taken the time to label what each piece means [1][3]. Instead of waiting for someone to label everything, unsupervised learning lets the computer discover hidden groups and patterns by itself [2].

What kinds of patterns does it find?

Unsupervised learning is really good at grouping similar things together, like sorting customers by shopping habits or finding groups of similar animals in pictures [5]. It can also find interesting connections in data that humans might not notice [3].

How is it different from other types of learning?

With supervised learning, you have to tell the computer "this is a cat" and "this is a dog" for thousands of pictures before it learns [1]. With unsupervised learning, you just hand it pictures and it figures out on its own which ones look similar [2][3].