PoD Can Mutate: A Simple Dynamic Directed Mutation Approach for Genetic Algorithms

The incorporation of contemporary directed mutation techniques into traditional genetic algorithms has shown significant potential and garnered many promising results. While such findings are important, the utility of directed mutation in a generic sense is limited by either a reliance on domain specific knowledge or an increase in complexity with respect to both parameter size and the functions required for random number generation. Thus, this paper presents a new approach to mutation that utilises a co-evolving set of simple directions for each chromosome that dictate the result of gene mutations. In addition to biasing mutation towards promising areas of the search space, the Pointed Directed (PoD) mutation technique requires no increase in a priori information about the domain, nor does it come at a significant cost to system complexity. Moreover, results show that PoD mutation offers significant advantages over gaussian mutation across a range of complex problem areas, maximising performance by utilising adaptation to tune both directions and emergent system momentum