Saward, Guy R.E.S. (1991) Interpreted dependency networks : a general framework for belief revision. PhD thesis, University of Warwick.

Preview

PDF WRAP_Theses_Saward_1991.pdf - Submitted Version - Requires a PDF viewer. Download (9Mb) | Preview

Request Changes to record.

Abstract

Belief Revision is a fundamental component of intelligent behaviour and is therefore an area of study within Artificial Intelligence (AI). In many cases the belief revision is the result of modifying a long term model of some domain by the addition of information specific to particular instances of a problem within that domain. This thesis argues the case for a Foundations approach to belief revision in which the level of belief in any proposition is supported by explicit justifications. Networks of such justifications can be used as long term knowledge stores thereby capturing the dependencies between different pieces of information.

Truth Maintenance Systems (TMSs), also known as Reason Maintenance Systems, are a class of programme that provides the functionality necessary to perform belief revision in just this way. However, each individual style of TMS contains embedded design decisions based on a particular problem solving domain and/or approach. An Interpreted Dependency Network (IDN) is the embodiment of the philosophy behind TMSs without the built-in assumptions. As such, IDNs allow for the easy specification of belief revision systems. Both the generality of IDNs and the ease of specification will be shown by rationally reconstructing existing approaches to TMSs.

This thesis provides a declarative semantics for IDNs along with general purpose algorithms for interpreting dependency networks. This means that IDNs have the necessary support to function as representations for long term knowledge. This is in contrast to the transitory nature of the information normally captured by TMSs.

The ability of IDNs to be used as knowledge representations is demonstrated by reconstructing several existing AI representation paradigms as IDNs. This also highlights the benefits of IDNs. For example, the declarative nature of the interpretation mechanism enables easy specification of the semantics of a representation while the explicit representation of dependencies enables a network to support several different modes of reasoning.

Item Type:	Thesis or Dissertation (PhD)
Subjects:	Q Science > QA Mathematics
Library of Congress Subject Headings (LCSH):	Artificial intelligence -- Computer programs, Truth maintenance systems
Official Date:	1991
Dates:	Date Event 1991 UNSPECIFIED
Institution:	University of Warwick
Theses Department:	Department of Computer Science
Thesis Type:	PhD
Publication Status:	Unpublished
Supervisor(s)/Advisor:	Cohn, A. G.
Sponsors:	Science and Engineering Research Council (Great Britain)
Extent:	x, 265, [15] leaves : illustrations
Language:	eng

Request changes or add full text files to a record

Repository staff actions (login required)

View Item

Downloads