Evolutionary algorithms (EAs) have been successfully applied to many real-world problems, often with success. However, the successful application of an EA to a problem typically requires the tuning of one-or-more parameters, which in itself is a non-trivial challenge. To overcome this problem, many researchers have proposed the use of self-adaptation, where the algorithm's various parameters are tuned during a run in response to feedback from the evolutionary search. This talk will examine one the concept of self-adaptation in EAs, with a particular focus on one method (self-adaptive neighbourhood search differential evolution). It will be shown that, while self-adaptation may offer some benefits in terms of easing the application of EAs to problems, the design of such methods is typically too ad hoc, and a more considered approach to their design should be taken.
Last modified: Tuesday, 09-Mar-2010 13:05:14 NZDT
This page is maintained by the seminar list administrator.