Fitness Costs of Mutation Rate Adaptation and its Application to Optimization of Dynamic Objective Functions

Evolutionary algorithms can be used to solve complex optimization tasks. However, adequate parameterization is crucial for efficient optimization. Evolutionary adaptation of mutation rates provides a solution to the problem of finding suitable mutation rate settings. However, evolution of low mutation rates may lead to premature convergence. In nature, mutation rate control coevolves with other functional units in a genome, and it is constrained because mutation rate control requires energy and resources. This principle can be captured by an abstract concept of fitness cost associated mutation rate adaptation, which can be generically applied in evolutionary algorithms. Application of this principle can be useful for addressing problems of premature convergence. This contribution explores applications of this concept within the context of dynamic fitness landscapes. It is shown that fitness costs for mutation rate adaptation is no less advantageous in dynamic fitness landscapes than in static ones, and that interesting synergies can arise in conjunction with dynamics in multimodal fitness landscapes