Accession Number : ADA192254
Title : Improving the Tools of Symbolic Learning.
Descriptive Note : Final rept. 1985-1987,
Corporate Author : PARIS-11 UNIV ORSAY (FRANCE) LAB DE RECHERCHE EN INFORMATIQUE
Personal Author(s) : Kodratoff, Yves
PDF Url : ADA192254
Report Date : 01 Sep 1987
Pagination or Media Count : 36
Abstract : Concepts relating to symbolic machine learning (ML) are discussed in this report. These concepts include knowledge representation, descriptive notations, and methods of generalization ML techniques have been applied to scene analysis through implementation of a system that learns features in order to recognize multi-font characters. Highlights of this research are discussed. In its first part, this paper presents some consequences of the choice of the definition of Generalization. It discusses the definitions based on deduction, versus those based on substitution. In its second part, it shows how symbolic computations are also able to take into account, at least partly, the noise most real-life data show. It discusses symbolic approaches to noise handling in Scene Analysis, rule learning, strategy learning and, finally, of the idea of polymorphic Version Space. Keywords: Deductive generalization, Generalization in an equational theory, Learning strategies, Polymorphy, Resistance to noise, Rule learning, Scene analysis, Version space.
Descriptors : *LEARNING MACHINES, COMPUTATIONS, FRANCE, FRENCH LANGUAGE, HANDLING, LEARNING, NOISE, RESISTANCE, STRATEGY, SUBSTITUTES, SYMBOLS, TOOLS, THEORY
Subject Categories : Human Factors Engineering & Man Machine System
Distribution Statement : APPROVED FOR PUBLIC RELEASE