Approximating Belief Functions in a Rule-Based System
Abstract
Rule-based expert systems have moved from a research activity in a small number of academic computer science departments to a growing commercial activity. This transition clearly indicates that the structure of a complex computer program enforced by a rule-based system (namely, the clear separation of the decision-making process, the inference engine, from the data on which the decisions are based, the rule base is a useful step in the evolution of programming strategies. At the same time there has been a growing recognition that in most decision-making situations the data (namely, the rule base and the initial evidence used to start the decision-making process) are not known with certainty and consequently the inference procedures used in traditional rule- based systems are inappropriate. Over the last decade a number of inference procedures which use various numerical representations of uncertainty have been developed for use in rule-based systems. However, for a variety of reasons (including the fact that there is little logical basis for the representations) none of them has been widely successful. This paper describes the current state of an ongoing research project which is attempting to use probability as the mechanism for representing uncertainty in a rule-based system.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 16, 1987
- Accession Number
- ADP005293
Entities
People
- William F. Eddy
Organizations
- Carnegie Mellon University