Meta-Lamarckian learning in three stage optimal memetic exploration

Abstract

Three Stage Optimal Memetic Exploration (3SOME) is a single-solution optimization algorithm where the coordinated action of three distinct operators progressively perturb the solution in order to progress towards the problem's optimum. In the fashion of Memetic Computing, 3SOME is designed as an organized structure where the three operators interact by means of a success/failure logic. This simple sequential structure is an initial example of Memetic Computing approach generated by means of a bottom-up logic. This paper compares the 3SOME structure with a popular adaptive technique for Memetic Algorithms, namely Meta-Lamarckian learning. The resulting algorithm, Meta-Lamarckian Three Stage Optimal Memetic Exploration (ML3SOME) is thus composed of the same three 3SOME operators but makes use a different coordination logic. Numerical results show that the adaptive technique is overall efficient also in this Memetic Computing context. However, while ML3SOME appears to be clearly better than 3SOME for low dimensionality values, its performance appears to suffer from the curse of dimensionality more than that of the original 3SOME structure.