Modelling selection intensity for linear cellular evolutionary algorithms.pdfVIP

Modeling Selection Intensity for Linear Cellular Evolutionary Algorithms 1 1 2 Mario Giacobini , Marco Tomassini , and Andrea Tettamanzi 1 Computer Science Institute, University of Lausanne, Switzerland {mario.giacobini, marco.tomassini}@unil.ch 2 Information Technologies Department, University of Milano, Italy andrea.tettamanzi@unimi.it Abstract. We present quantitative models for the selection pressure on cellular evolutionary algorithms structured as a ring of cells. We obtain results for synchronous and asynchronous cell update policies. Theoreti- cal results are in agreement with experimental values and show that the selection intensity can be controlled by using different update methods. 1 Introduction Cellular evolutionary algorithms (cEAs) are an example of spatially structured evolving populations that is often used in optimization and other applications [1]. The structure may be an arbitrary graph, but more commonly it is a one- dimensional or two-dimensional grid. This kind of evolutionary algorithm has become popular because it is easy to implement on parallel hardware of the SIMD or MIMD type. Although SIMD machines have almost disappeared except for special purpose computing, cEAs can still be very conveniently implemented on computer clusters with excellent performance gains. However, what really matters is the model, not its implementation. Thus, in this work we will focus on cEA models and on their properties without worrying about implementation issues. The theory of cEAs is relatively underdeveloped, although several results have been published on selection pressure and convergence speed. Sarma and De