Optimal sampling and speed-up for genetic algorithms on the sampled OneMax problem
24 June 2003Yu, T.-L., Goldberg, D. E., Sastry, K. (2003). Proceedings of the Genetic and Evolutionary Computation Conference. 1554—1565. [Full paper - PDF] [Full paper - PS].
Abstract:
- This paper investigates the optimal sampling and the speed-up obtained through sampling for the sampled OneMax problem. Theoretical and experimental analyses are given for three different population-sizing models: the decision-making model, the gambler’s ruin model, and the fixed population-sizing model. The results suggest that, when the desired solution quality is fixed to a high value, the decision-making model prefers a large sampling size, the fixed population-sizing model prefers a small sampling size, and the sampling makes no difference for the gambler’s ruin model. The speed-up obtained by sampling is then empirically verified.
Posted in Conference Proceedings, Principled Efficiency Enhancement Techniques, Publications |
Trackback |
Top Of Page
Related Posts:
- Don’t evaluate, inherit
- Using previous models to bias structural learning in the hierarchical BOA
- Probabilistic model building and competent genetic programming
- Towards billion bit optimization via efficient genetic algorithms
- Toward routine billion-variable optimization using genetic algorithms
Comments are closed.
