Inductive Genetic Programming and Superposition of Fitness Landscapes

This paper presents an approach to improving the performance of evolutionary algorithms. The evolutionary search e®ort is distributed among cooperating subpopulations that correspond to the substructures of the ̄tness landscape. The idea is to create such subpopulations that °ow easily on the simple substructures of the complex ̄tness landscape structure. We claim that the search on a complex ̄tness landscape is facilitated if properly integrated with search on its simple components. This evolutionary structured search is applied for solving hard inductive learning tasks. The performance observed while inducing regular grammars from sets of boolean strings demonstrated that the approach mitigates the search di±culties.