The case for lower probabilities as measures of uncertainty
- Oak Ridge National Lab., TN (United States)
- Tennessee Univ., Knoxville, TN (United States). Dept. of Mathematics
This paper presents the case for using lower probabilities as measures of uncertainty in expert systems. A debate has raged within the artificial intelligence community for years about how to represent uncertainty in expert systems. Several camps have emerged. One camp has focused on developing alternatives to probability theory, such as certainty factors, fuzzy sets, and endorsements. A second camp has focused on retrofitting classical, additive probability, for example, by developing a cautious approach to probabilistic reasoning and interpreting probability within a possible worlds framework. This paper falls into a third camp, which encompasses generalizations of probability theory. The most discussed generalization is known as Dempster-Shafer Theory, which is based on the combined work of Dempster and Shafer. Lower probabilities are actually a substantial generalization of DST. This paper has two parts. The first presents the definitions of lower probabilities, DST, and additive probability. This section includes a discussion of capacities, the most general type of uncertainty measure. The purpose of this section is to show the differences among the uncertainty measures.
- Research Organization:
- Oak Ridge National Lab., TN (United States)
- Sponsoring Organization:
- EPRI; Electric Power Research Inst., Palo Alto, CA (United States)
- DOE Contract Number:
- AC05-84OR21400
- OSTI ID:
- 5305154
- Report Number(s):
- CONF-9109121-1; ON: DE91017870
- Country of Publication:
- United States
- Language:
- English
Similar Records
Decaying confidence functions for aging knowledge in expert systems
Incorporating ''fuzzy'' data and logical relations in the design of expert systems for nuclear reactors