Construction and Local Computation Aspects of Network Belief Functions.

摘要

Belief function methodology extends Bayesian reasoning about uncertainty to situations where the available evidence does not support a full probabilistic specification. A network belief function is constructed by combining independent component belief functions representing logical and probabilistic knowledge assessed locally across the network. Computational methods for combining (propagating and fusing) the local knowledge assessments are sketched. Examples are introduced without detailed formulas, including the Kalman filter, fault trees, and sampling inference. Keywords: Bayes Theorem, Functions, Reprints. (cd)