Latin Hypercube Sampling in Bayesian Networks

Wepropose a schemefor producingLatin hypercube samples that can enhanceany of the existing sampling algorithms in Bayesiannetworks. Wetest this scheme in combinationwith the likelihood weightingalgorithm andshowthat it can lead to a significant improvement in the convergence rate. While performance of sampiing algorithms in general dependson the numerical properties of a network, in our experimentsLatin hypercube sampling performed always better than randomsampling. In some cases we observed as much as an order of magnitudeimprovementin convergence rates. Wediscusspractical issues related to storage requirements of Latin hypercube sample generation and propose a low-storage, anytime cascaded version of Latin hypercube sampling that introduces a minimal performanceloss comparedto the original scheme.1

• Type: Conference
• Year: 2000
• Where: FLAIRS
• Authors: Jian Cheng, Marek J. Druzdzel
• Keywords: Artificial Intelligence, FLAIRS 2000, Hypercube, Hypercube Sample, Latin Hypercube