A study on the computational efficiency of Baldwinian evolution

Memetic algorithms are search methods coupling population-based evolution and individual learning, and are attracting growing attention in the recent two decades. As computational cost is usually an essential factor of concern, there has been increasing research on optimizing and adapting the frequency and budget of individual learning, in order to achieve efficient search. However, most of current research concentrates on Lamarckian scenario. In this paper, we investigated into what Baldwinian learning brings to the evolution, from the view of computational efficiency. it is revealed in the work that in Baldwinian evolution, the learning effort hardly pushes the search going further, but just maintains a certain level of potential to reach good solutions, thus diversity. Baldwinian learning brings hardly any improvement in computational efficiency, and variation of learning budget may break the potential and even reduce the search efficiency.