Session:

TUT - Tutorials

Title:

Spatially Structured EAs

Authors:

Marco Tomassini

Abstract:

The topological structure of evolving populations has a marked influence on the dynamical aspects of evolutionary algorithms. In this tutorial we will examine several possibilities for the spatial structuring of populations. In particular, island and diffusion models will be discussed in detail, with reference to genetic algorithms, genetic programming and other EA types. It will be shown how diversity can be better preserved in spatially structured populations, and the effect of topology on selection pressure in cellular EAs will be discussed. The time dimension of the operations will also be taken into account, introducing synchronous and asynchronous spatially structured EA models. Finally, although the emphasis will be on models rather than implementations, several possible ways of implementing the models on parallel hardware will be described, together with experimental results on benchmarks and other problems.