Cultural Evolution of Ensemble Learning for Problem Solving

As the complexity of problem spaces increases in terms of their inherent dimensionality and the interaction between the dimensions, it becomes increasing harder to rely on a small subset of dimensions to guide an evolutionary optimizer to a solution. In this paper we integrate a popular machine learning technique, ensemble learning, into the evolutionary process using the cultural algorithms framework. Each of five different knowledge sources in the cultural algorithms belief space is viewed as part of an ensemble where each interacts with the other to control the exploration of the population. We apply the prototype to the design of a tension compression spring and demonstrate the advantages of evolving such ensembles for optimization applications.