Randomness can be exploited to update control parameters (F and CR) into adaptive DE algorithms. There are several instances for sequential DE algorithms but so few examples in the distributed/parallel case.
Actually I have done some work on that. See
Matthieu Weber, Ferrante Neri, Ville Tirronen: A study on scale factor/crossover interaction in distributed differential evolution. Artif. Intell. Rev. 39(3): 195-224 (2013)
Matthieu Weber, Ferrante Neri, Ville Tirronen: A study on scale factor in distributed differential evolution. Inf. Sci. 181(12): 2488-2511 (2011)
Matthieu Weber, Ville Tirronen, Ferrante Neri: Scale factor inheritance mechanism in distributed differential evolution. Soft Comput. 14(11): 1187-1207 (2010)
Dear Ferrante, thank you for your answer. Actually, I and my colleagues know and appreciate your works. A couple of rows in our bibliographies are usually assigned to your papers, indeed :). By the way, there is not so much study on this issue. Am I wrong? Any suggestion?
Q: Why random "adaptation"? In my humble experience adaptation has always been deterministic. Could this be an application for fuzzy control? I use a fuzzy controller inspired by the work of Lee & Takagi to adapt population size, crossover rate, and mutation rate in my evolution system (which includes DE). System can be described as evolution algorithms + data structures.
Dynamic Control of Genetic Algorithms using Fuzzy Logic Techniques (1993)
by Michael A. Lee , Hideyuki Takagi
Venue: PROCEEDINGS OF THE FIFTH INTERNATIONAL CONFERENCE ON GENETIC ALGORITHMS
I've done some work on the adaptation of control parameters for the DE as well. Maybe you would like to check these papers:
Rodrigo Pedrosa C Silva, Rodolfo A Lopes, Alan R R Freitas, Frederico G Guimaraes: Performance Comparison of Parameter Variation Operators in Self-Adaptive Differential Evolution Algorithms. Neural Networks, Brazilian Symposium on 01/2012; pp.148-153
Rodrigo César Pedrosa Silva, Rodolfo Ayala Lopes, Frederico Gadelha Guimarães:
Self-adaptive mutation in the differential evolution.
13th Annual Genetic and Evolutionary Computation Conference, GECCO 2011, Proceedings, Dublin, Ireland, July 12-16, 2011; 01/2011
@Keenan: Thank you for your partecipation. By "random" adaptation I mean the way you update the algorithm's parameters. One can imagine to design a rule to decide when and how changing the parameter values. In both cases, it is possible to use either a deterministic or random approach. You could refer to the papers previously suggested by Ferrante Neri for having some instance of a random approach.
@Pedrosa Silva: Thanks to you too. My question is about distributed DE and your works seem to regard sequential algorithms. Anyway, thank you again.
If you are looking for the "soft-spot" of parameter setting of a given metaheuristic on a given problem, perhaps you should try Bonesa: http://sourceforge.net/projects/tuning/
But I am not sure it can handle distributed implementations.
I think Bonesa can do that (and more). Selmar Smit ([email protected]) can tell you everything about it.
@dos Santos,@Eiben: I didn't know Bonesa. At first glance, Bonesa is helpful for parameter tuning. My question is quite different: I would like to have some information about the presence in literature of adaptive distributed DE algorithms with their own control parameters not fixed before execution but updated at run time by sampling a random distribution. Thus, I'd rather be interested in models than tools. Anyway, thank you for your comments.
I would like to recommend you the following article:
http://www.sciencedirect.com/science/article/pii/S0378779615001947
- Badges
- Science method