Biomimetic Representation with Enzyme Genetic Programming

Abstract

The standard parse tree representation of genetic programming, whilst a good choice from a generative viewpoint, does not capture the variational demands of evolution. This paper addresses the issue of whether representations in genetic programming might be improved by mimicry of biological behaviours, particularly those thought to be important in the evolution of metabolic pathways, the 'computa- tional' structures of the cell. This issue is broached through a presentation of enzyme genetic programming, a form of genetic programming which uses a biomimetic representation. Evaluation upon problems in combinational logic design does not show any significant performance advantage over other approaches, though does demonstrate a number of interesting behaviours including the preclusion of bloat.