| Evolutionary computation is a heuristic optimization technique inspired by nature. The optimization problem is analogous to the natural environment, candidate solutions become individuals in the population, and relative solution quality determines individual fitness. Evolution, acting through selection and reproduction, directs the search to find an optimal solution.;Biologists have studied "Darwin's finches" on the Galapagos archipelago and found that the El Nino effect, which causes periodic drought-induced scarcity, and monsoon-providing abundance, is driving a noticeable genotypic and phenotypic adaptation. I utilize a periodic selection event, and population growth parameter controls, to simulate this natural phenomenon. I demonstrate that these mechanisms provide additional efficiency and robustness to the evolutionary search.;Sewall Wright's Shifting Balance Theory indirectly tells us that the population size of a sub-species is proportional to its fitness: more-fit sub-populations will have more individuals and be in an exploitation mode, while less-fit sub-populations will have fewer individuals and be in an exploration mode. I use this as inspiration to allocate offspring to different sub-populations (species) in unequal amounts, through the selection process. The less-fit species will win fewer offspring, causing them to explore the search space through weaker selection pressures and random genetic drift. I show that this is a beneficial mechanism for multi-modal search, granting the ability to escape from local optima, and also as a means of automatically searching more densely in more promising locations. That is, this mechanism helps to achieve an automatic balance between exploration and exploitation.;These two inspirations are implemented in the Speciating Selection Event algorithm. I focus on understanding the behavior of the algorithm, and how effectively it is able to search the problem space. To do this, I make use of conceptually simple mathematical functions, rather than a real-world or engineering task.;In this work I also analyze the behavior of one of the canonical algorithms in the field of evolutionary computation, furthering the understanding of the feedback mechanism through which the problem environment and selection method effect the evolutionary search. This understanding is also put to use in the analysis of the Speciating Selection Event algorithm.;... |
