Instance-based learning by searching

Instance based learning (IBL) methods fascinate with their conceptual simplicity. Usually, IBL systems center on (a subset of) given “training” instances. Restricting the acquisition of instances to given training instances is known to cause difficulties and limitations. We propose to overcome these limitations by searching for suitable instances. We employ a genetic algorithm to conduct such an intricate search. We demonstrate the viability of this approach in connection with instance based concept learning