Illinois Genetic Algorithms Laboratory

Miller, B. L., & Goldberg, D. E. (1996)
TR No.: 96005 | Download PDF | Download PS

Abstract:
Abstract:

In many real-world environments, a genetic algorithm designer is often faced with choosing the best fitness function from a range of possibilities. Fitness functions differ primarily based upon the speed, accuracy, and cost of a fitness evaluation. An important type of fitness function is the \textit{sampling fitness function}, which utilizes sampling in order to reduce the noise of fitness evaluations. The accuracy and speed of a sampling fitness function are directly related to the \textit{sample size}, which is the number of samples used by the sampling fitness function to evaluate an individual chromosome. The optimal sample size denotes the sample size that maximizes the performance of a genetic algorithm within a fixed time bound. In this paper, a domain independent lower bound of the optimal sample size is derived, and a sample size pruning method is described.