Definition
A computer program, loosely inspired by the brain, that learns patterns and makes predictions from many examples instead of hand-written rules.
At a glance
- Learns from examples rather than being programmed, so more good data makes it better.
- Built from simple units called neurons stacked in layers, with a tunable weight on each connection.
- Training is costly and data-hungry; using the trained model is fast and cheap.
- Powers product recommendations, fraud detection, forecasting, image analysis, and chatbots.
How it works
Data enters one end, passes through hidden layers of neurons that each transform it, and an answer comes out[2]. Each connection carries a number called a weight[1]. Nobody writes the rules: you show it thousands of labeled examples, it guesses, sees how wrong it was, and nudges its weights to improve[3].
Why it matters
These tools turn your past data into predictions. Retailers recommend products and forecast demand; banks flag fraud in real time; healthcare and manufacturing analyze images and predict failures; chatbots run on them too[4].
Watch out
Answers are only as good as the data, and the model can be a black box that is hard to explain. That matters for regulated decisions like loans. For most owners, buying these capabilities from vendors beats building from scratch.
Bottom line
A pattern-learner that turns past data into predictions; most businesses should use existing products rather than build one.