ABSTRACT
Traditional evolutionary algorithms (EAs) are powerful problem solvers that have several fixed parameters which require tuning. An increasing body of evidence suggests that the optimal values of some, if not all, EA parameters change during the course of executing an evolutionary run. This paper investigates the potential benefits of dynamic parameters by applying a Meta-EA to evolving optimal dynamic parameter values for population size, offspring size, n in n-point crossover, Gaussian mutation's step size, bit flip mutation's mutation rate, parent selection tournament size, and survivor selection tournament size.
Each parameter was optimized both as the only dynamic parameter, and with all parameters dynamic. The most effective two parameters when acting independently were also allowed to optimize in tandem. The results were compared with a Meta-EA tuned EA using static parameters on the DTrap, NK, Rastrigin, and Rosenbrock benchmark problems. Results support that all tested parameters have the potential to improve solution fitness by changing dynamically, and using multiple dynamic parameters was more effective than using each independently.
- J. Cook and D. Tauritz. An Exploration into Dynamic Population Sizing. In Proceedings of GECCO 2010. Portland, Oregon, USA, 2010. Google ScholarDigital Library
- A. Eiben, M. Schut, and A. deWilde. Is Self-Adaptation of Selection Pressure and Population Size Possible? In Proceedings of PPSN IX, pages 900--909, 2006. Google ScholarDigital Library
- J. Gomez. Self Adaptation of Operator Rates in Evolutionary Algorithms. In Proceedings of GECCO 2010, pages 162--173. Springer Berlin, Heidelberg, 2004.Google ScholarCross Ref
- A. Nwamba and D. Tauritz. Futility-Based Offspring Sizing. In Proceedings of GECCO 2009, pages 1873--1874, 2009. Google ScholarDigital Library
- G. Papa. Parameter-less Evolutionary Search. In Proceedings of GECCO 2008, pages 1133--1134. Atlanta, GA, USA, 2008. Google ScholarDigital Library
- J. Smith. Modeling GAs with Self-Adaptive Mutation Rates. In Proceedings of GECCO 2001, pages 599--606, 2001.Google Scholar
- E. Smorodkina and D. Tauritz. Toward Automating EA Configuration: the Parent Selection Stage. In Proceedings of IEEE CEC, pages 63--70, 2007.Google ScholarCross Ref
- F. Vafaee, W. Xiao, P. Nelson, and C. Zhou. Adaptively Evolving Probabilities of Genetic Operators. Machine Learning and Applications, pages 292--299, 2008. Google ScholarDigital Library
Index Terms
- Meta-evolved empirical evidence of the effectiveness of dynamic parameters
Recommendations
An exploration into dynamic population sizing
GECCO '10: Proceedings of the 12th annual conference on Genetic and evolutionary computationTraditional evolutionary algorithms are powerful problem solvers that have several fixed parameters which require prior specification. Determining good values for any of these parameters can be difficult, as these parameters are generally very sensitive,...
Levy distributed parameter control in differential evolution for numerical optimization
Differential evolution (DE) algorithm is a population based stochastic search technique widely applied in scientific and engineering fields for global optimization over real parameter space. The performance of DE algorithm highly depends on the ...
To explore or to exploit: An entropy-driven approach for evolutionary algorithms
An evolutionary algorithm is an optimization process comprising two important aspects: exploration discovers potential offspring in new search regions; and exploitation utilizes promising solutions already identified. Intelligent balance between these ...
Comments