Files in this item



application/pdf8815328.pdf (5MB)Restricted to U of Illinois
(no description provided)PDF


Title:An Inductive Engine for the Acquisition of Temporal Knowledge
Author(s):Chen, Kaihu
Doctoral Committee Chair(s):Michalski, Ryszard S.,
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Computer Science
Abstract:The ability to predict the likely occurrence of certain events in the future permits one to make plans in advance in order to achieve a goal. This capability can be acquired empirically by discovering that certain temporal patterns repeat unerringly. Previous work in time-series analysis allows one to make quantitative predictions on the likely values of certain linear variables. Since certain types of knowledge are better expressed in symbolic form, making qualitative predictions based on symbolic representations requires a different approach. This thesis describes a domain-independent method, called TIM (Time-based Inductive Machine), for discovering symbolic temporal patterns from observations using the technique of inductive inference. The problem of discovering temporal patterns from observations can be viewed as a problem in concept acquisition, where the target event to be predicted can be viewed as a class designator, and the "causes" to be discovered can be viewed as the hypothesized descriptions for that class. Using this approach, the "causes" of the target event can be discovered with the application of generalization and specialization operators to partial hypotheses in an orderly way. Three issues that concern the utility of the method in real world domains are addressed. First, a representation based on a modification of first-order predicate logic for expressing temporal concepts/observations is proposed. Generalization/specialization operators and inductive heuristics based on this representation are also given. Second, due to the prevalence of uncertainty in real-world domains, the method is designed to learn probabilistic concepts, as well as concepts with fuzzy boundaries. And finally, the method is designed to learn incrementally in order to offset the penalties incurred by the use of this powerful representation.
Issue Date:1988
Description:139 p.
Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 1988.
Other Identifier(s):(UMI)AAI8815328
Date Available in IDEALS:2014-12-15
Date Deposited:1988

This item appears in the following Collection(s)

Item Statistics