Glossary

[Manipulation | Learning from Data | Glossary | References ]

Glossary

This is the glossary of terms used in the Software for manipulating belief networks and Software for fitting belief functions to data pages.

Key to terms used in features and lacks lists.

Bayesian Networks

(also Bayes Nets ) purely probabilistic models which use an acyclic directed graph to represent conditional independence assumptions among variable and probabilities as the primary representation of uncertainty. (Pearl[1988] and Lauritzen and Spiegelhalter [1988])

Belief Functions

Supports belief functions (Dempster-Shafer evidential reasoning) as a representation of uncertainty (this feature almost always shows up in the lacks list.) The Graphical-Belief web site contains a good discussion of the distinctions between belief functions a probabilities.

Belief Networks

Catch-all term used to include graphical belief functions, Bayesian networks, influence diagrams, and probabilistic graphical models.

Chain Graphs

Supports graphs with a mixture of directed and undirected edges. (Lauritzen and Wermuth [1989])

Common Lisp the Language, Version 1

(Guy Steele) Published in 1984, this was the de facto standard for Common Lisp before the ANSI standard committee started meeting. Indicates that the software may require some adaptation for recent lisps.

Common Lisp the Language, Version 2

(Guy Steele) Published in 1990, this revision tracked most of the changes in the Common Lisp standard of the ANSI standard committee. This version is available over the web: Steele[1990] Indicates that this program may not be compatible with older lisps.

Dynamic Models

Contains facilities dynamic creation/modification of models. This is espeically useful in temporal reasoning models.

Explanation

Contains facilities for explaning the behavior of models.

Gaussian Variables

Indicates that the model handles continuous variables with a Gaussian (normal) distribution.

GUI

Has (lacks) a Graphical User Interface.

Graphical Belief Function

Model expressible as the sum of independent belief functions. The factorization structure of the model is represented by a hypergraph. As probabilities are a special case of belief functions, this includes probabilistic graphical models as a special case. (Almond [1995]).

Influence Diagram

Models which include both probabilities and utilities. These include Bayesian networks as a special case. (Howard and Matheson [1981])

Knowledge Based Model Construction

Facilities for helping the user build complex models from small pieces or rules. (Almond, Bradshaw and Madigan [1994])

Local Expression Language

Program can represent and exploit (during inference) of noisy ors, asymmetries, contingencies, and other special case models. ( Techincal Report by Bruce D'Ambrosio on available on this subject.)

Networks

Indicates that the program allows any network conforming expressible as an acyclic directed graph.

Model Criticism and Learning

Indicates the ability to update the structure of the model in the presence of new data. (Cowell, Dawid and Spiegelhalter [1993])

Parameter Uncertainty

Allows the user to specify a law for a parameter of a probability or belief function distribution and second order models. (Note: A law is the distribution of a parameter of a distribution. Thanks to Steffen Lauritzen for suggesting this terminology.) (Almond[1995] and Spiegelhalter and Lauritzen [1990])

Probabilities

Supports only probabilities as representations of uncertainty, not belief functions (Usually based on Lauritzen and Spiegelhalter [1988] or Pearl[1988] formalism.)

Probabilities/Belief Functions

Supports both probabilities and belief functions as representations of uncertainty.

Probabilistic Graphical Models

Similar to a Bayesian network but expressed on a undirected graph.

* Possibilities

Supports possiblity theory (Zadeh [1978], Dubois and Prade [1988a, b]) as an alternative representation of uncertainty.

Standard Rulebase

Uses the proposed Bayesian Network Standard format.

Tree Shaped Bayesian Networks

Restricted to models which can be expressed as trees (no undirected cycles.)

Valuations

Supports the full valuation system of Shenoy and Shafer [1990] for generic inference.

Utilities

Includes a representation of preference as well as uncertainty. This implies that the package supports influence diagrams and not just Bayesian networks.

[Manipulation | Learning from Data | Glossary | References ]