Expert Systems and Statistical Expertise. Part 1. Statistical Expert Systems
Abstract
An expert system is a computer program at the level of a human expert in a complex but narrow field. Two types of expert systems which involve statistical expertise are statistical consulting programs and programs which find patterns in databases. Consulting programs can now be built quickly using programming tools. Twosamp, a consultant for advising on the analysis of univariate two-sample problems, was constructed in less than a week using one such tool, EMYCIN, Systems like Twosamp cannot model the initial stages of a consulting session, but can provide expert assistance on a class of models has been selected. Consultants and pattern-finders can assist statisticians in building models, generating hypotheses, and maintaining complex databases. As well, they can serve as laboratories for experimentation with statistical strategies. Expert systems also use statistical expertise. Most systems include mechanisms for reasoning under uncertainty. Methods under investigation include fuzzy logic, Dempster-Shafer theory, Bayesian analysis and various ad hoc methods. Learning systems use statistics to infer inductive rules. As well, statistical reasoning will be used to evaluate the performance of expert systems.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1985
- Accession Number
- ADA159503
Entities
People
- N. S. Altman
Organizations
- Stanford University