The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #94134

Title:Rudy - a Tool for Defining and Evaluating Qualitative Combining Functions in a Belief Network
Date: 1994
Abstract:Belief networks are a method for reasoning under uncertainty. The uncertainty inherent in statements about an application domain are expressed in probabilities. Belief networks, which are basically directed acyclic graphs, are also a way of modeling an expert's knowledge about a particular domain. The nodes in such a graph represent propositions about the domain and the arcs denote dependencies between them. Given the initial probabilities of some nodes an inference algorithm calculates the conditional probabilities of the remaining nodes in the network. Another way of representing uncertainty is to use qualitative values instead of probabilities. In this case an expert is asked to express his belief in a proposition with a certain range of values, e.g. from strongly believed to not believed or from confirmed to disconfirmed. Using this method one also has to provide qualitative combining functions which calculate the belief in a proposition given the value of belief of its connected nodes. Rather than using a global function that does the task one lets the expert define local combining rules for each node in the network. This particular approach is followed in this thesis. The aim was to develop a graphical tool that allows experts firstly to model an application domain n a belief network and secondly to define the combining rules for each node. Thirdly the expert can run the model, i.e. let the rules fire and propagate the levels of belief of some initial propositions. The resulting tool dynamically generates a rule based system by translating the user defined combining rules into CLIPS rules. The tool is also flexible in the way that it can easily be applied to models other than belief networks.

[Search These Pages] [DAI Home Page] [Comment]