PII: 0888-613X(88)90170-3


346 A b s t r a c t s  

assumes marginal independence between inputs. This demonstrates that the importance of 
updating formulas can outweigh that o f  prior assumptions. Thus, when UISs are judged 
by their final accuracy after optimization, completely different results are obtained than 
when they are judged by whether or not their prior assumptions are perfectly satisfied. 

Structuring Causal Tree Models with Continuous Variables 

L e i  X u  
Department o f  Automation, Tsinghua University, Beijing, China 
J u d e a  P e a r l  
Cognitive Systems Laboratory, Computer Science Department, 
UCLA, Los Angeles, California 90024-1600 
This paper considers the problem o f  invoking auxiliary, unobservable variables to 
facilitate the structuring o f  causal tree models for a given set o f  continuous variables. 
Paralleling Pearl's 1986 treatment o f  bivalued variables, it is shown that if a collection of 
coupled variables are governed by a joint normal distribution and a tree-structured 
representation exists, then both the topology and all internal relationships of the tree can 
be uncovered by observing pairwise dependencies among the observed variables (i.e., the 
leaves o f  the tree). Furthermore, the conditions for normally distributed variables are less 
restrictive than tho3e governing bivalued variables. The result extends the applications of 
causal tree models that were found useful in evidential reasoning tasks. 

Address correspondence to L. Xu. 

Implementing Evidential Reasoning in Expert Systems 

J o h n  Y e n  
USC/Information Sciences Institute, 4676 Admiralty Way, Marina 
del Re),, California 90292 
The Dempster-Shafer theory has been extended recently for its application to expert 
systems. However, implementing the extended D-S reasoning model in rule-based 
systems greatly complicates the task o f  generating informative explanations. By 
implementing GERTIS, a prototype system for diagnosing rheumatoid arthritis, it is 
shown that two kinds of knowledge are essential for explanation generation: (1) 
taxonomic class relationships between hypotheses and (2) pointers to the rules that 
significantly contribute to belief in the hypothesis. As a result, the knowledge represented 
in GERTIS is richer and more complex than that of conventional rule-based systems. 
GERTIS not only demonstrates the feasibility o f  rule-based evidential reasoning systems, 
but also suggests ways to generate better explanations and to explicitly represent various 
useful relationships among hypotheses and rules. 

Can Evidence Be Combined in the Dempster-Shafer Theory? 

J o h n  Y e n  

USC/lnformation Sciences Institute, 4676 Admiralty Way, Marina 
del Rey, California 90292