Premium
A theory of schema labelling
Author(s) -
Havens William
Publication year - 1985
Publication title -
computational intelligence
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.353
H-Index - 52
eISSN - 1467-8640
pISSN - 0824-7935
DOI - 10.1111/j.1467-8640.1985.tb00065.x
Subject(s) - computer science , schema (genetic algorithms) , theoretical computer science , knowledge base , star schema , information schema , graph , artificial intelligence , database schema , semi structured model , data mining , information retrieval , database design
Schema labelling is a knowledge organization theory for recognition systems. In this theory, recognition constructs symbolic network descriptions of sensory data in terms of a stored knowledge base of schemas. A schema knowledge representation is formally specified. Each schema represents a class of objects by specifying class membership as a set of constraints on other classes. The constraints are based on two orthogonal relationships in knowledge organization: composition and specialization. Complex objects have internal structure which can be represented as composition constraints among their parts. Likewise, specialization constraints segment classes into subclasses by type. The description produced by schema labelling is a network consistency graph. The nodes of the graph are schema instances derived from the knowledge base. The domain for each instance is its finite set of subclasses and the network constraints are the constraints defined between schemas. Constraints are propagated in the network using an arc consistency algorithm mat has been adapted to schema representations. The constructed network description makes explicit the objects recognized in the data and their relationships. The graph is hierarchical, providing a description at multiple levels of detail. A hypothetical scene analysis system is used to illustrate the methodology.