Comparative study of effects of fitness landscape changes in open-ended evolutionary simulations and in Genetic Programming

The biological world where natural selection takes place is a world constantly affected by external physical phenomena, whether cyclical and regular, such as the rotation of the earth, or punctual such as when a meteorite strikes the earth. It is recognized that these phenomena affect the evolution of life, but their interaction with natural selection have not yet been fully explored.

This thesis studies the effects of environmental fluctuations via evolutionary simulations. In particular, we propose to study them through simulations of Genetic Programming as well as Artificial Life evolving virtual ecosystems.

We first present the history of natural selection as well as artificial life and genetic programming, focusing on the role of environmental fluctuations in these three fields of research.

We then examine the effects of such fluctuations on virtual ecosystems. We create a new virtual ecosystem which we call HetCA, that is based on cellular automata with heterogeneous transition rules. In this simulation, we test effects of these fluctuations on the evolutionary progress as well as on the level of selection and show that the type of fluctuation determine the level of selection.

Finally, we study the effects of such fluctuations in Genetic Programming, firstly through an extension of Random Interleaved Sampling and then by creating a new method of Genetic Programming which we call Wave. We note that, on the studied problems, Wave is a very competitive method compared to a selection of benchmarks including non-evolutionary computation based optimization methods.