Data-driven Constructive Induction in the Learnable Evolution Model
Janusz Wojtusiak · 2008
The learnable evolution model (LEM) is a non-Darwinian evolutionary computation method which applies symbolic machine learning to guide the evolutionary optimization process. This paper investigates application of data-driven constructive induction to automatically improve representation spaces in LEM. This includes investigation of methods for modifying representation spaces and methods for creating new candidate solutions from hypotheses learned in the modified spaces. Experimental results indicate that LEM equipped with constructive induction outperforms LEM working only in the original representation spaces.