The present dissertation would have never been achieved without the support of a variety of people. I am particularly indebted to Jose Antonio Lozano, my thesis supervisor. Undoubtedly, his wise, while at the same time friendly guidance throughout these years has made me grow a lot. Jose Antonio, thanks for your invaluable help and patience. I am also grateful to Pedro Larrañaga, Iñaki Inza, Alex Mendiburu, Endika Bengoetxea and the rest of my colleagues at the Intelligent Systems Group. Their encouragement and advice have been decisive as well. I would like to make special mention of my lab

I present a new estimation-of-distribution approach to program evolution where distributions are not estimated over the entire space of programs. Rather, a novel representationbuilding procedure that exploits domain knowledge is used to dynamically select program subspaces for estimation over. This leads to a system of demes consisting of alternative representations (i.e. program subspaces) that are maintained simultaneously and managed by the overall system. Metaoptimizing semantic evolutionary search (MOSES), a program evolution system based on this approach, is described, and its representation-building subcomponent is analyzed in depth. Experimental results are also provided for the overall MOSES procedure that demonstrate good scalability.

Abstract. We present N-gram GP, an estimation of distribution algorithm for the evolution of linear computer programs. The algorithm learns and samples a joint probability distribution of triplets of instructions (or 3-grams) at the same time as it is learning and sampling a program length distribution. We have tested N-gram GP on symbolic regressions problems where the target function is a polynomial of up to degree 12 and lawn-mower problems with lawn sizes of up to 12 × 12. Results show that the algorithm is effective and scales better on these problems than either linear GP or simple stochastic hill-climbing. 1