Abstract

I generalise the arguments of [Chow & Liu 1968] to show that a Bayesian network satisfying some arbitrary constraint that best approximates a probability distribution is one for which mutual information weight is maximised. I give a practical procedure for nding an approximation network and evaluate its application on a range of data sets. Articial intelligence requires the ability to reach conclusions that may be far from certain. For example an expert system for medical diagnosis may be given the symptoms of some patient and asked to provide a diagnosis | even though the background knowledge and symptom information may not be enough to determine for sure which problem actually besets the patient. Probability theory provides a plausible model for reasoning under uncertainty, since one would expect a diagnosis to be relatively probable, given the symptoms. This paper addresses practical issues to do with the implementation of probabilistic reasoning. The plan is rst to discuss...

Citations