An Algorithm for Fast Optimal Latin Hypercube Design of Experiments

[Felipe A. C. Viana]

Dear all,

Here it is a reference on analysis of uncertainty structures of
surrogate models:

F. A. C. Viana, G. Venter and V. Balabanov, "An Algorithm for Fast
Optimal Latin Hypercube Design of Experiments," International Journal
for Numerical Methods in Engineering, available online, 2009 (DOI:
10.1002/nme.2750).

This paper presents the translational propagation algorithm, a new
method for obtaining optimal or near optimal Latin hypercube designs
(LHDs) without using formal optimization. The procedure requires
minimal computational effort with results virtually provided in real
time. The algorithm exploits patterns of point locations for optimal
LHDs based on the PHIp criterion (a variation of the maximum distance
criterion). Small building blocks, consisting of one or more points
each, are used to recreate these patterns by simple translation in the
hyperspace. Monte Carlo simulations were used to evaluate the
performance of the new algorithm for different design configurations
where both the dimensionality and the point density were studied. The
proposed algorithm was also compared against three formal optimization
approaches (namely random search, genetic algorithm, and enhanced
stochastic evolutionary algorithm). It was found that (i) the
distribution of the PHIp values tends to lower values as the
dimensionality is increased and (ii) the proposed translational
propagation algorithm represents a computationally attractive strategy
to obtain near optimum LHDs up to medium dimensions.


You can find more about it online:
http://fchegury.googlepages.com/publications

All the best,
Felipe A. C. Viana