Selection Procedures for Module Discovery: Exploring Evolutionary Algorithms for Cognitive Science

Evolutionary algorithms are playing an increasingly important role as search methods in cognitive science domains. In this study, methodological issues in the use of evolutionary algorithms were investigated via simulations in which procedures were systematically varied to modify the selection pressures on populations of evolving agents. Traditional roulette wheel, tournament, and variations of these selection algorithms were compared on the “needle-in-ahaystack” problem developed by Hinton and Nowlan in their 1987 study of the Baldwin effect. The task is an important one for cognitive science, as it demonstrates the power of learning as a local search technique in smoothing a fitness landscape that lacks gradient information. One aspect that has continued to foster interest in the problem is the observation of residual learning ability in simulated populations even after long periods of time. Effective evolutionary algorithms balance their search effort between broad exploration of the search space and in-depth exploitation of promising solutions already found. Issues discussed include the differential effects of rank and proportional selection, the tradeoff between migration of populations towards good solutions and maintenance of diversity, and the development of measures that illustrate how each selection algorithm affects the search process over generations. We show that both roulette wheel and tournament algorithms can be modified to appropriately balance search between exploration and exploitation, and effectively eliminate residual learning in this problem. Introduction: EC and Cognitive Science Evolutionary computation (EC) is increasingly used in cognitive science, both for evolving cognitive models and for modeling evolutionary processes. Many algorithms use evolutionary search in one form or another. No single search algorithm will be optimal for all tasks (a thesis colloquially known as “no free lunch”, Wolpert & Macready, 1996). In any simulation study, characteristics of the task need to be taken into account in the selection of algorithms. However, to many cognitive science researchers it is not clear which aspects of tasks are important in the design of a search process, and what properties of evolutionary search algorithms need to be taken into account to select an appropriate design. This study is part of a wider program of research whose goal is to enhance the effective use of evolutionary computation techniques in cognitive science domains. This program involves assessing the performance of popular evolutionary algorithms on tasks of interest to cognitive scientists. Current areas in cognitive science that are utilizing EC methods include the direct modeling of evolutionary processes, such as the role of learning in evolution, learning as a local search technique in a genetic algorithm, the evolution of modularity, the evolution of cooperation, and the evolution and learnability of language (e.g., see the biennial “Evolution of Language” conferences, or the Evolutionary Computation “Special Issue on EC and Cognitive Science”, Wiles & Hallinan, 2001). Other domains use evolutionary algorithms for optimization, for example, testing theories of infant development; modeling populations of individuals engaged in cognitive tasks; testing outcomes following damage in neural network models; and exploring the range of behaviors in a dynamic model of an artificial language learning task. In all of the cognitive science domains mentioned, evolutionary algorithms have been tested on specific problems, but little work has been done at a methodological level to characterize the nature of the tasks per se, and the way in which they interact with the evolutionary algorithms chosen. Many factors affect the performance of evolutionary algorithms, including the choice of fitness function, representation of the genome, population size, selection technique, and genetic operators.