PII: 0888-613X(88)90140-5


334 Abstracts 

UCLA, Los Angeles, California 90024-1600 
An inductive logic can be formulated in which the elements are not propositions or 
probability distributions, but information systems. The logic is complete for information 
systems with binary hypotheses, that is, it applies to all such systems. It is not complete 
for information systems with more than two hypotheses but applies to a subset of such 
systems. The logic is inductive in that conclusions are more informative than premises. 
Inferences using the formalism have a strong justification in terms of the expected value 
of the derived information system. 

Explanation of Probabilistic Inference for Decision Support 
Systems 

C h r i s t o p h e r  E l s a e s s e r  
MITRE Washington Artificial _Intelligence Center and Department 
o f  Engineering and Public Policy, Carnegie-Mellon University, 
Pittsburgh, Pennsylvania 15213 
This paper reports work in progress on an explanation facility for Bayesian conditioning 
aimed at improving user acceptance of probability-based decision support systems. 
Design of the facility, which appears to be reasonably domain-independent, is based on 
an information processing model that accounts for both biased and normative behavior in 
reasoning about conditional evidence. Preliminary results indicate that the facility is both 
acceptable to naive users and effective in improving understanding of Bayesian 
conditioning. 

Automated Generation of Connectionist Expert Systems for 
Problems Involving Noise and Redundancy 

S t e p h e n  I. G a l l a n t  
College o f  Computer Science, Northeastern University, Boston, 
Massachusetts, 02115 
When creating an expert system, the most difficult and expensive task is that of 
constructing a knowledge base. This is particularly true if the problem involves noisy data 
and redundant measurements. This paper shows how to modify the MACIE process for 
generating connectionist expert systems from training examples so that it can accommo- 
date noisy and redundant data. The basic idea is to dynamically generate appropriate 
training examples by constructing both a " d e e p "  model and a noise model for the 
underlying problem. The use of winner-take-all groups of variables is also discussed. 

These techniques are illustrated with a small example that would be very difficult for 
standard expert system approaches. 

A Measure-Free Approach to Conditioning 

I. R. G o o d m a n  
Command and Control Department, Code 421, Naval Ocean 
Systems Center, San Diego, California 92152