On the Utility of Redundant Encodings in Mutation-Based Evolutionary Search

Abstract

A number of recent works in the evolutionary computation field have suggested that introducing large amounts of genetic redundancy may increase the evolvability of a population in an evolutionary algorithm. These works have variously claimed that the reliability of the search, the final fitness achieved, the ability to cope with changing environments, and the robustness to high mutation rates, may all be improved by employing this strategy. In this paper we dispute some of these claims, arguing that adding random redundancy cannot be generally useful for optimization purposes. By way of example we report on experiments where a proposed neutral encoding scheme (based on random Boolean networks) is compared to a direct encoding in two mutation-only EAs, at various mutation rates. Our findings show that with the appropriate choice of per-bit mutation rate, the evolvability of populations using the direct encoding is no less than with the redundant one.