A comparison of probabilistic and fuzzy inference for expert systems
Date
1990
Authors
Akin, Garth Stuart
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Investigators differ over approaches to managing inexact inference for expert systems, particularly concerning. probabilistic and fuzzy inference. Analysis of the literature suggests this is due partly to glossing distinctions between types of inexactness, and partly to lack of a commonly accepted theoretical framework.
Probabilistic and fuzzy inference are compared by presenting a formal language, and interpreting it in a metric lattice. It is shown that if the lattice is modular, the truth-value assignments for Standard Uncertainty Logic (Gaines) can be derived; if the lattice is Boolean the value-assignments for Unconditional Probability Logic (Rescher) can be derived; if the lattice is a chain, the value assignments for a Fuzzy Logic (Zadeh) can be derived; if the lattice is a Boolean chain the value-assignments for Sentential Calculus can be derived. It is shown how the set of models can be implemented as inference procedures for an expert system shell. A comparison of the probabilistic and fuzzy inference procedures shows that their results are practically the same.