Critical Remarks on Belief Updating in Bayesian Networks

D. Wu (Canada)


Bayesian networks, probabilistic reasoning, belief updat ing, local propagation


There exists a number of problems in the traditional method for belief updating. First, it is generally believed that the junction tree propagation (JTP) method cannot compute p(X|e) when X is not contained in a node of the junction tree. Secondly, the local propagation procedure has to be applied whenever new evidence is observed. Many researchers have attempted to solve the first problem. Contrary to common belief, in this paper we show that one can in fact easily compute p(X|e) by the standard JTP method for any X. We also show that it is not neces sary to repeat the local propagation procedure for process ing new evidence. More importantly, perhaps, we suggest a more efficient method for belief updating. Our method requires to compute the marginals of the individual nodes in the junction tree only once.

Important Links:

Go Back