Define instance-based AI methods.

Instance-based AI methods, referred to as lazy learning methods, are non-parametric techniques that focus on local inference rather than global modeling. These methods derive their predictions based on previously encountered similar cases, operating as needed. An example of this approach is the nearest-neighbor methods, which are flexible and can adjust their complexity according to the available data. They also have the capability for universal approximation, meaning they can handle various tasks as long as suitable data structures are available. Instance-based methods rely on memorizing single instances, which helps them identify out-of-sample instances based on similarity to experienced data, and they can effectively deal with catastrophic forgetting in continual learning scenarios by retaining relevant prior data points. This principle also enables them to respond to distributional changes effectively, although the choice of representation greatly influences their generalization ability【1】.