Knowledge incorporation through lifetime learning

Evolutionary computation is known to require long computation time for large problems. This chapter examines the possibility of improving the evolution process by incorporating domain-specific knowledge into evolutionary computation through lifetime learning. Different approaches to combining lifetime learning and evolution are compared. While the Lamarckian approach is able to speed up the evolution process and improve the solution quality, the Baldwinian approach is found to be inefficient. Through empirical analysis, it is conjectured that the inefficiency of the Baldwinian approach is due to the difficulties for genetic operations to produce the genotypic changes that match the phenotypic changes obtained by learning. This suggests that combining evolutionary computation inattentively with any learning method available is not a proper way to construct hybrid algorithms; rather, the correlation between the genetic operations and learning should be carefully considered.