Abstract:A probability-updating method in probabilistic expert systems is considered in this paper based on the minimum discrimination information. Here, newly acquired information is taken as the latest true marginal probabilities, not as newly observed data with the same weight as previous data. Posterior probabilities are obtained by up- dating prior probabilities subject to the latest true marginals. To apply to probabilistic expert systems, we extend Ku and Kullback(1968)'s the minimum discrimination infor- mation method for saturated models to log-linear models, discuss localization of global updating, and show that Deming and Stephan's iterative procedure can also be used to find the posterior probabilities. Our updating method can also be used to handle uncertain evidences in probabilistic expert systems. Updating and acquisition of knowledge in probabilistic expert systems have discussed by many investigators, such as Spiegelhalter and Lauritzen(1990) and Spiegelhalter, Dawid, Lauritzen and Cowell(1993). In their discussions, new acquired information is taken as data, and then probabilities in a expert system are updated with the new data and prior proba- bilities or previous data by using Bayesian method or the likelihood method. However, with the development of new tools and methods for acquiring information, the newly acquired information becomes more and more precise and believable such that it must be taken as the latest true probabilities to replace the prior probabilities, not as the new data with the same weight as the previous data. Usually such information is hard to be complete or global for a large system. Therefore, we ought to deal with the incomplete information as the latest true marginal probabilities, and we update the prior probabilities subject to the specified marginal probability restriction such that discrimination information between the prior and the posterior probabilities is minimized. Ku and Kullback(1968) and Ireland and Kullback(1968) discussed an updating method for saturated models of contingency tables based on the minimum discrimination informa- tion. In this paper, we shall extend their method to log-linear models, discuss localization of global updating, and we shall show that Deming and Stephan's iterative procedure can also be used to find the posterior probabilities. Our updating method can also be used to handle uncertain evidences in probabilistic expert systems. The uncertain evidences are specified

Updating Probabilistic Knowledge Using Imprecise and Uncertain Evidence

Conditional Dempster-Shafer Theory for Uncertain Knowledge Updating

Conditional Evidence Theory and Its Application in Knowledge Discovery

Belief measures conditioned on random set

Generalized Jeffrey'S Rule of Conditioning and Evidence Combining Rule for A Priori Probabilistic Knowledge in Conditional Evidence Theory

MINIMUM INFORMATION UPDATING WITH SPECIFIED MARGINALS IN PROBABILISTIC EXPERT SYSTEMS

The generalized combination of evidence involved a priori knowledge

Belief Update in Bayesian Networks Using Uncertain Evidence

New Probabilistic Transformation of Imprecise Belief Structure

Sampling-Based Adaptive Bayesian Quadrature for Probabilistic Model Updating

Efficient Updating of Probabilistic Approximations with Incremental Objects

An improved approach for incomplete information modeling in the evidence theory and its application in classification

Judicious Judgment Meets Unsettling Updating: Dilation, Sure Loss, and Simpson's Paradox

Bayesian Network Revision with Probabilistic Constraints

Bayesian Update with Information Quality under the Framework of Evidence Theory

A new probabilitic transformation in evidence theory

Bayesian Updating Method Under Random and Interval Hybrid Uncertainties

Multiscale probability transformation of basic probability assignment

Bayesian Network Reasoning With Uncertain Evidences

Getting Wiser from Multiple Data: Probabilistic Updating according to Jeffrey and Pearl

Hierarchical Proportional Redistribution for Bba Approximation