Coevolutionary and Coadaptive Systems: Papers from the AAAI Fall Symposium
Mitchell A. Potter and R. Paul Wiegand, Cochairs
Coevolutionary and coadaptive systems are techniques in which multiple interacting elements of a system are learned concurrently in relationship to one another. Such methods include both competitive and cooperative systems. While competitive systems tend to involve evaluating a candidate solution using a coadapting opponent to measure its performance, cooperative systems tend to involve evaluating a collaborative assembly of multiple coadapting components. These systems offer promise for applications involving interactive domains. Interest in coevolutionary and coadaptive systems has intensified in recent years as new theory has developed. Using tools such as evolutionary game theory and order theory, as well as tools drawn from multiobjective optimization, to describe the dynamics of coevolutionary algorithms and the structure of coevolutionary problems, analysis underscores the need to understand the relationship between the underlying problems one wishes to solve and the nature of the applied algorithms. This analysis has lead to more realistic expectations of the potential of coevolutionary and coadaptive systems, as well as clarifications of their goals. We have begun to see the design and implementation of more useful algorithms as a result.
As new applications of coevolutionary and coadaptive systems emerge in areas such as multiobjective optimization and robotics, they continue to reveal their advantages and drawbacks, exposing many interesting research issues. For example, questions surrounding problem decomposition and role assignment, among others, continue to present interesting challenges for researchers in the field. These and other research issues provide a compelling motivation for a symposium focused on contemporary topics in the field