Defining Implicit Objective Functions for Design Problems
The ability of evolutionary algorithms and related search techniques to explore a varied space of solutions with efficiency and often surprising innovation makes them useful tools for design. This typically requires the explicit definition of a goal or objective function and so has been ideally suited to engineering optimisation tasks. For many design problems however, and particularly for those of great complexity, it is difficult to specify such a goal in advance. Design and creativity themselves, particularly in a social context, are often seen as processes of guided, but open exploration. Steels has shown that effective languages can be generated without an external measure of quality by allowing robots to speak and evaluate each other in an environment. Such approaches have been incorporated into genetic algorithms by allowing the objective to change over time.
Publication:Proceedings of GECCO '07: Genetic And Evolutionary Computation Conference | full text (PDF)
Year: 2007