Tuesday, January 25, 2005
Niching methods, Mahfoud's population-sizing, and scalability of GAs
While working on papers for GECCO, I was re-reading Sam Mahfoud's paper titled "population sizing for Sharing methods". The population-sizing estimate though derived for fitness sharing, also holds for other niching methods such as restricted tournament selection as well.
Since niching mechanism is one of the key components of hierarchical and multiobjective genetic algorithms, the population-sizing paper deserves renewed attention. As Martin and I recently found out the hard way—to which prof. Goldberg replied: 'if you had talked to me before doing the experiments, I would have told you so' :-) —the scalability of the niching method played an important role in the overall scalability of multiobjective hBOA.
Since niching mechanism is one of the key components of hierarchical and multiobjective genetic algorithms, the population-sizing paper deserves renewed attention. As Martin and I recently found out the hard way—to which prof. Goldberg replied: 'if you had talked to me before doing the experiments, I would have told you so' :-) —the scalability of the niching method played an important role in the overall scalability of multiobjective hBOA.
Comments:
<< Home
I can only agree with the point you made and I think we can take it even a bit further---Population sizing of all shapes and sizes deserves more attention and many researchers in GEC don't realize this. For anyone interested in looking into this topic, I'd suggest starting with Goldberg & Rudnick (1991) and Harik et al. (1997).
Post a Comment
<< Home
