Measuring the Hardness of Stochastic Sampling on Bayesian Networks with Deterministic Causalities: the kTest
Haohai Yu, Robert van Engelen
Abstract:
Approximate Bayesian inference is NPhard. Dagum and Luby defined the Local Variance Bound (LVB) to measure the approximation hardness of Bayesian inference on Bayesian networks, assuming the networks model strictly positive joint probability distributions, i.e. zero probabilities are not permitted. This paper introduces the ktest to measure the approximation hardness of inference on Bayesian networks with deterministic causalities in the probability distribution, i.e. when zero conditional probabilities are permitted. Approximation by stochastic sampling is a widelyused inference method that is known to suffer from inefficiencies due to sample rejection. The ktest predicts when rejection rates of stochastic sampling a Bayesian network will be low, modest, high, or when sampling is intractable.
Keywords:
Pages: 786795
PS Link:
PDF Link: /papers/11/p786yu.pdf
BibTex:
@INPROCEEDINGS{Yu11,
AUTHOR = "Haohai Yu
and Robert van Engelen",
TITLE = "Measuring the Hardness of Stochastic Sampling on Bayesian Networks with Deterministic Causalities: the kTest",
BOOKTITLE = "Proceedings of the TwentySeventh Conference Annual Conference on Uncertainty in Artificial Intelligence (UAI11)",
PUBLISHER = "AUAI Press",
ADDRESS = "Corvallis, Oregon",
YEAR = "2011",
PAGES = "786795"
}

