Hierarchical Evidence and Belief Functions
This work tackles the problem of representing and propagating uncertain evidence in automated reasoning systems, but it appears incremental as it builds on existing D/S theory methods without introducing a new paradigm.
The paper addresses the challenge of transforming rules with attached beliefs into joint belief functions for propagation in Dempster/Shafer theory, demonstrating through examples that multiple joint functions can be consistent with a given rule set and that different rule representations yield varying beliefs on consequents.
Dempster/Shafer (D/S) theory has been advocated as a way of representing incompleteness of evidence in a system's knowledge base. Methods now exist for propagating beliefs through chains of inference. This paper discusses how rules with attached beliefs, a common representation for knowledge in automated reasoning systems, can be transformed into the joint belief functions required by propagation algorithms. A rule is taken as defining a conditional belief function on the consequent given the antecedents. It is demonstrated by example that different joint belief functions may be consistent with a given set of rules. Moreover, different representations of the same rules may yield different beliefs on the consequent hypotheses.