A machine learning method where the generalization of training data is delayed until a query is made to the system.
For instance, a recommendation system that only generates recommendations when a user makes a query, rather than proactively suggesting items before the user has even asked for them.