question on function topology
Aziz A Lookman
I have begun to use GAs for function optimization. My initial reading
suggests that GAs are at an advantage over other methods, such as
derivative-based ones, when the function landscape is "rugged". Also,
which variant of the GA to use seems to depend on function topology.
I would very much appreciate any pointers on how to evaluate function
topology to get a sense of how many valleys and peaks are there, and
whether the highest point is at the top of a rounded hill, or on the top
of a colum.
Thank you all for your input.
***********************************************************************
Aziz A. Lookman, Ph.D. Candidate
R. Paul Wiegand
I cannot speak specifically to your topology questions, but I can
provide a view basic starting points for multi-modal research.
There's more out there than this, but a few of good papers on finding
multiple peaks from a function are as follows:
Spears, W. M. (1994). "Simple subpopulation schemes." In Sebald &
Fogel (Eds.) _Proceedings of the Third Conference on Evolutionary
Programming, pp. 297-307
Goldberg, D. E., Deb, K., & Horn, J. (1992) "Massive multimodality,
deception, and genetic algorithms". In _Proceedings of the Parallel
Problem Solving from Nature Conference, pp. 37-46.
Deb, K. and Goldberg, D. E. (1989). "An investigation of niche and
species formation in genetic function optimization". In _Proceedings
of the Third International Conference on Genetic Algorithms, pp.
42-50.
I don't believe these papers address your issues directly, but perhaps
they can service you as starters, or direction finders. My
understanding is that all three of these papers are based on research
techniques for finding multiple optima simultaneously via a variety of
techniques including speciation and fitness sharing.
On Wed, 26 Nov 1997 11:59:41 -0500, Aziz A Lookman <aa...@CMU.EDU>
wrote: