Adaptive mutation for semi-separable problems

Abstract

In this paper we introduce a new mutation heuristic in an attempt to better match genetic algorithms and the geography of search spaces. This is achieved by varying the mutation rate across the genotype to more rapidly search those areas that are currently believed to be having the greatest detrimental impact on the phenotype fitness. The new adaptive mutation operator is shown to be efficient in two applications: fault diagnosis in distributed and multiprocessor systems and the classical traveling salesman problem. We believe that the proposed, adaptive, mutation operator is the first step in realizing a new class of adaptive genetic operators for use with a distinct, but common, subset of real world applications.