The one-inclusion graph algorithm is near-optimal for the prediction model of learning

TitleThe one-inclusion graph algorithm is near-optimal for the prediction model of learning
Publication TypeJournal Articles
Year of Publication2001
AuthorsLi Y, Long PM, Srinivasan A
JournalIEEE Transactions on Information Theory
Volume47
Issue3
Pagination1257 - 1261
Date Published2001/03//
ISBN Number0018-9448
KeywordsComputer science, concept class, general-purpose algorithm, graph theory, learning prediction model, learning systems, near-optimal algorithm, Neural networks, one-inclusion graph algorithm, optimisation, Prediction algorithms, prediction theory, Predictive models, probability, Probability distribution, Upper bound, Vapnik-Chervonenkis dimension, Virtual colonoscopy
Abstract

Haussler, Littlestone and Warmuth (1994) described a general-purpose algorithm for learning according to the prediction model, and proved an upper bound on the probability that their algorithm makes a mistake in terms of the number of examples seen and the Vapnik-Chervonenkis (VC) dimension of the concept class being learned. We show that their bound is within a factor of 1+o(1) of the best possible such bound for any algorithm

DOI10.1109/18.915700