Introduction
Some data never stop arriving: clicks on a website, card transactions, sensor readings, stock prices, social-media posts. Retraining a model from scratch every few minutes would be far too slow and expensive. Online learning models instead update themselves continuously, one example (or one small batch) at a time.
Picture a weather forecaster who adjusts their predictions a little every day as new observations come in, instead of re-studying 50 years of records each morning.
This topic covers how streaming learning works in practice and how to handle the biggest challenge: the world changing under your model.