Proximity and priority: applying a gene expression algorithm to the Traveling Salesperson Problem

Abstract

In this paper we describe an environment for evolutionary computation that supports the movement of information from genome to phenotype with the possibility of one or more intermediate transformations. Our notion of a phenotype is more than a simple alternate representation of the binary genome. The construction of a phenotype is sufficiently different from the genome as to require its generation by a procedure that we call a gene expression algorithm. We discuss various reasons why benefits should accrue when combining gene expression algorithms with conventional genetic algorithms and illustrate these ideas with an algorithm to generate approximate solutions to the traveling salesperson problem. As in most genetic algorithms dealing with the TSP we run into the problem of an appropriate crossover operation for the strings that specify a permutation. To handle this issue we introduce a novel genome representation that admits a natural crossover operation and produces a permutation vector as an intermediate representation.