BaRT (Bayesian Reasoning Tool) Manual Preliminary Version 2.0
Abstract
Many real world problems are associated with uncertainty; the evidence people observe which helps them to reason about some goal event is almost always uncertain and incomplete. Still, people make judgements based on this uncertain and incomplete evidence. These uncertain evidences can be combined in various ways to find the validity or strength of a hypothesis and Bayesian probability theory is a normative theory that allows one to reason about and combine uncertainties. Pearl has devised a way to represent, reason about and combine uncertain evidences in a way that conforms to the tenets of probability theory, but avoids the disadvantages usually associated with probabilistic computations of belief. BaRT is a Bayesian Reasoning Tool which implements Pearl's ideas. It has been implemented as an AI programming environment which can perform classification problem solving, and it has been used to classify ships . In BaRT, a classification problem is represented as a network of hypotheses. The belief in each value of each hypothesis can change as new evidence lends support to (or takes support away from) certain values of the hypothesis. Section 2 provides an overview of the theoretical background for this work. Section 3 explains how to use BaRT and provides an example. Sections 4 and 5 provide details concerning the implementation of this system. This manuals describes a preliminary version of a system which is under development. Later versions of BaRT will have greater capabilities, so any of the functions and capabilities described here are subject to change. Keywords: PCL(Portable Common Loops).
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 22, 1988
- Accession Number
- ADA197829
Entities
People
- C. L. Ramsey
- L. B. Booker
- N. Hota
Organizations
- United States Naval Research Laboratory