Version space controlled genetic algorithms (VGA)

Abstract

Demonstrates how the traditional genetic algorithm can be augmented by incorporating domain knowledge in the form of a version space into the structure. This hybrid inductive learning system is designed to handle problems in concept learning using the version space to control the search process that is performed by the genetic algorithm. In this hybrid system a new class of schemata is present called hyperschema. A theorem for hyperschema analogous to that for traditional schema is presented. This theorem demonstrates how the addition of domain knowledge in the form of a version space allow the hybrid system to exploit schemata of higher order and defining length via a hitchhiking effect. A prototype program that implements this approach is briefly described in pseudocode and the actual system is used to solve a problem in concept learning posed by Nilsson (1987). >