The Influence of Learning in the Optimization of Royal Road Functions

Authors

Abstract

We will apply a hybrid Genetic Algorithm with learning to a class of functions widely studied in Evolutionary Computation, known as Royal Road functions. Individual learning, in this context, is usually implemented as a local search algorithm which is applied to some individuals of the population. Local search is the computational analogue to phenotypic plasticity in biological systems. Two different local search procedures will be used, RMHC and DHC. Experiments performed prove that they have a major impact in the results obtained. We will also consider two different learning strategies: Lamarckian learning and the Baldwin effect. We show that, for this problem and with a correct learning procedure, Lamarckian learning outperforms the Baldwin effect.