On learning more concepts

Technical Report

Öffentlich Deposited

Citeable URL: https://ir.library.oregonstate.edu/concern/technical_reports/ng451r82x

Descriptions

Attribute Name	Values
Creator	Almuallim, Hussein Dietterich, Thomas G.
Abstract	The coverage of a learning algorithm is the number of concepts that can be learned by that algorithm from samples of a given size. This paper asks whether good learning algorithms can be designed by maximizing their coverage. The paper extends a previous upper bound on the coverage of any Boolean concept learning algorithm and describes two algorithms-Multi-Balls and Large-Ball-whose coverage approaches this upper bound. Experimental measurement of the coverage of the ID3 and FRINGE algorithms shows that their coverage is far below this bound. Further analysis of Large-Ball shows that although it learns many concepts, these do not seem to be very interesting concepts. Hence, coverage maximization a.lone does not appear to yield practically useful learning algorithms. The paper concludes with a definition of coverage within a bias, which suggests a way that coverage maximization could be applied to strengthen weak preference biases. Keywords: inductive learning, concept coverage, theoretical analysis.
Resource Type	Technical Report
Date Issued	1992
Academic Affiliation	Computer Science
Series	Technical report (Oregon State University. Department of Computer Science)
Urheberrechts-Erklärung	In Copyright
Publisher	Oregon State University. Department of Computer Science
Peer Reviewed	No
Language	English [eng]

Miniaturansicht	Titel	Hochladedatum	Sichtbarkeit	Aktionen
	Almuallim_Dietterich_92_30_02.pdf	2022-12-29	Öffentlich	Herunterladen