 |
    |
Adaptive Sampling for Noisy Problems
Erick Cantú-Paz
Center for Applied Scientific Computing, Lawrence Livermore National Laboratory, Livermore, CA 94551cantupaz@llnl.gov
Abstract. The usual approach to deal with noise present in many real-world optimization problems is to take an arbitrary number of samples of the objective function and use the sample average as an estimate of the true objective value. The number of samples is typically chosen arbitrarily and remains constant for the entire optimization process. This paper studies an adaptive sampling technique that varies the number of samples based on the uncertainty of deciding between two individuals. Experiments demonstrate the effect of adaptive sampling on the final solution quality reached by a genetic algorithm and the computational cost required to find the solution. The results suggest that the adaptive technique can effectively eliminate the need to set the sample size a priori, but in many cases it requires high computational costs.
LNCS 3102, p. 947 ff.
lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2004