Influence Diagrams: Automated Analysis with Dynamic Programming
Abstract
The major goals of this thesis were to develop a user friendly software package for processing influence diagrams, and to implement in software the extensions necessary for dynamic programming without special action or knowledge on the part of the user. The final goal was to demonstrate the efficiency of the dynamic programming techniques by applying them to several example problems. A software package, AFids (AFIT influence diagram system) was developed. The system is capable of performance equivalent to the current state of the art in commercial influence diagram software. AFids incorporates the basic influence diagram operations, the separable value function extensions, and an algorithm to automatically solve any properly formed influence diagram. Separation of the value function is automatic and requires no action or special knowledge on the part of the user beyond representing the value function explicitly as a sum or product. The software uses menus, data entry screens, and graphics to provide an effective and friendly user interface. Several extensions to influence diagram theory were implemented in the AFids package including the concepts of value rounding and outcome limiting to control the combinatorial explosion encountered when processing chains of deterministic nodes. The system is cost free and available in either source or compiled form for government users. There are no restrictions on use or distribution except for commercial use. The software runs on MS-DOS compatible microcomputers and was programmed in Turbo Pascal.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1988
- Accession Number
- ADA203045
Entities
People
- Christopher T. Baron
Organizations
- Air Force Institute of Technology