In this lecture, we move from talking about multi-objective genetic algorithms to distributed and parallel genetic algorithms. The discussion of distributed genetic algorithms (DGA) as a metapopulation shows us how, in principle, small populations undergoing evolutionary processes can collectively move through "adaptive valleys" to explore much better than a monolithic population. This idea that distributed subpopulations that each make use of genetic drift as a mechanism for exploration is precisely the shifting-balance theory of Sewall Wright, a theory which has not been extremely useful in biology but may be a perfect fit for evolutionary computing in a distributed setting. We then discuss how our insights from multiobjective and distributed GA's can be used to reinvigorate approaches for single-objective, single-processor GA's and so-called "multimodal optimization" (which will be described in the next lecture).
Whiteboard notes for this lecture can be found at:
No comments:
Post a Comment