Research output

Conference Paper (in Formal Publication) ()

Partial Structure Learning by Subset Walsh Transform

Citation
Christie LA, Lonie D & McCall J (2013) Partial Structure Learning by Subset Walsh Transform In: Jin Y, Thomas SA (ed.) Computational Intelligence (UKCI), 2013 13th UK Workshop on, Piscataway, NJ, USA: IEEE. 13th UK Workshop on Computational Intelligence (UKCI), 2013, 9.9.2013 - 11.9.2013, Guildford, pp. 128-135.

Abstract
Estimation of distribution algorithms (EDAs) use structure learning to build a statistical model of good solutions discovered so far, in an effort to discover better solutions. The non-zero coefficients of the Walsh transform produce a hypergraph representation of structure of a binary fitness function; however, computation of all Walsh coefficients requires exhaustive evaluation of the search space. In this paper, we propose a stochastic method of determining Walsh coefficients for hyperedges contained within the selected subset of the variables (complete local structure). This method also detects parts of hyperedges which cut the boundary of the selected variable set (partial structure), which may be used to incrementally build an approximation of the problem hypergraph.

StatusPublished
EditorJin Y, Thomas SA
AuthorsChristie Lee A, Lonie David, McCall John
Publication date31/10/2013
Date of public distribution09/2013
URLhttps://openair.rgu.ac.uk/handle/10059/1387
PublisherIEEE
Place of publicationPiscataway, NJ, USA
ISSN of series 2162-7657
ISBN 978-1-4799-1567-5
eISBN 978-1-4799-1568-2
LanguageEnglish
© University of Stirling FK9 4LA Scotland UK • Telephone +44 1786 473171 • Scottish Charity No SC011159
My Portal