Impact of Regulatory Genes on Optimization Behavior.

Daniel Ashlock and Wendy Ashlock
Submitted to CEC 2012

Abstract PDF eprint

In nature, regulatory genes determine which part of an organism's gnome is expressed. In this study a simple regulatory mechanism is used to modify linear representations. The regulatory mechanism substantially enhances exploration at the expense of exploitation. For complex, polymodal fitness landscapes the modification yields a substantial improvement in performance. A negative control example is designed that demonstrates the technique yields a remarkable degradation of performance on a unimodal optimization problem designed to interact poorly with the technique. Analysis shows that the regulatory mechanism creates the potential for insertion and deletion mutations within the linear representation. These mutations have the effect of substantially increasing the number of genomes one mutation away from any given genome. This has the effect of decreasing the diameter of any search space where they regulatory technique is implemented.