Supervised learning
Learning from examples where each input comes with the correct answer.
Think of it like
Like studying with an answer key — you check each attempt against the right answer and adjust.
Example
Training a model to detect fraud from thousands of transactions each already marked “fraud” or “legit.”
How it actually works
Supervised learning is the classic setup: given labeled input-output pairs, learn a mapping that predicts the output for new inputs. It’s the most direct and well-understood paradigm, powering classification and regression everywhere. Its bottleneck is labels — they’re expensive to produce and their quality caps performance, which is exactly why self-supervised pretraining became so attractive.
For product teams
The dependable default when you have labeled data; its cost lives in getting those labels.
For engineers
Learn input→output from labeled pairs (classification/regression); performance bounded by label quantity and quality.
Related
- Labeling — Needs labeling.
- Unsupervised learning — Contrast with unsupervised learning.
- Reinforcement learning — Also contrast with reinforcement learning.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome