Coevolution is a promising problem solving technique that can be applied in interactive domains. This includes problems in which candidate solutions are evaluated using tests and problems in which multiple components are to be co-adapted. By adaptively choosing which interactions to evaluate, coevolution can provide an evaluation function that is not only efficient but may also be more accurate than a fixed, user-defined evaluation function.

As a result of this capability, as well as their often complicated dynamics, coevolutionary algorithms have historically provided researchers with much hope and consternation. Intuition about their potential abounded, and early interest in coevolution was stimulated by a number of successful results. Despite this promising start, however, coevolution has been employed in only a very limited number of practical applications so far.

Our theoretical understanding of coevolution has meanwhile developed substantially. Current views of coevolution place a strong emphasis on understanding the relationship between the underlying problem that one wishes to solve and the nature of the coevolutionary algorithm that is being applied. This focus leads to a more realistic expectation of the potential of coevolutionary algorithms, and clarifies the goals of coevolutionary algorithms for different problems.

At last year's GECCO, an informal and unofficial meeting was held where terminology in the field of coevolutionary computation was discussed. This very lively discussion was felt by all participants to be a valuable event, and has resulted in a much clearer picture of how researchers in the field currently use terminology and where more consistent terminology could be established. Indeed, this discussion led to the establishment of a Coevolution Wiki as a means of maintaining public discourse on the subject. Based on the success of this meeting, there will be a similar forum discussion this year that is open to any interested attendants as part of the GECCO Workshop Program.

AIMS AND FORMAT

The 2005 coevolution discussion forum will be an open meeting in the form of a two-hour forum discussion. The general point of the forum is to foster an open discussion concerning relevant issues in coevolutionary computation with active audience participation. The specific aims of this meeting are twofold:

• Determine the current potential for application of coevolution;
• Identify what issues must be addressed in order move from the current theoretical understanding of coevolution towards a practical set of problem solving techniques.

To this end, five topics that cover the main open issues in coevolution have been identified to provide a structure for the discussion:

• Terminology: results of last year's discussion and new input;
• Identification and description of problem classes that coevolution can in principle address, and specific subclasses for which coevolution is or is not expected to be effective;
• Appropriate solution concepts for different problem classes;
• The match between specific coevolutionary algorithms and problem classes;
• Applying coevolution successfully.

An invited panel of active coevolution researchers will serve to ensure that a discussion will be initiated. The discussion will be moderated by a member of the panel.

CONTACT INFORMATION