Semi-supervised learning is a type of machine learning that’s a mix of supervised and unsupervised learning.
Imagine you’re learning to identify different breeds of dogs. You have a few pictures of dogs where a teacher has told you what breed each one is. You also have a lot more pictures where you don’t know the breed. You start by studying the pictures with labels and learn some traits of different breeds. Then, you look at the unlabeled pictures and try to guess the breed based on what you learned. You might even find some new traits that help you recognize breeds better. That’s a bit like semi-supervised learning.
In semi-supervised learning, a computer program, or model, is given a small amount of labeled data (like the pictures of dogs where you know the breed) and a large amount of unlabeled data (the pictures where you don’t know the breed). The model learns from the labeled data first, then uses what it learned to find patterns and make educated guesses about the unlabeled data.
Semi-supervised learning can be very useful when we have a lot of data, but only a small amount of it is labeled, because labeling data can be time-consuming and expensive. It’s used in many areas of artificial intelligence, including image recognition and natural language processing.
« Back to Glossary Index