Skip navigation
Skip navigation

A scalable parallel FEM surface fitting algorithm for data mining

Christen, Peter; Hegland, Markus; Roberts, Stephen; Altas, Irfan

Description

The development of automatic techniques to process and detect patterns in very large data sets is a major task in data mining. An essential subtask is the interpolation of surfaces, which can be done with multivariate regression. Thin plate splines provide a very good method to determine an approximating surface. Unfortunately, obtaining standard thin plate splines requires the solution of a dense linear system of order n, where n is the number of observations. Thus, standard thin plate splines...[Show more]

dc.contributor.authorChristen, Peter
dc.contributor.authorHegland, Markus
dc.contributor.authorRoberts, Stephen
dc.contributor.authorAltas, Irfan
dc.date.accessioned2003-07-03
dc.date.accessioned2004-05-19T12:19:22Z
dc.date.accessioned2011-01-05T08:37:56Z
dc.date.available2004-05-19T12:19:22Z
dc.date.available2011-01-05T08:37:56Z
dc.date.created2001
dc.identifier.urihttp://hdl.handle.net/1885/40729
dc.identifier.urihttp://digitalcollections.anu.edu.au/handle/1885/40729
dc.description.abstractThe development of automatic techniques to process and detect patterns in very large data sets is a major task in data mining. An essential subtask is the interpolation of surfaces, which can be done with multivariate regression. Thin plate splines provide a very good method to determine an approximating surface. Unfortunately, obtaining standard thin plate splines requires the solution of a dense linear system of order n, where n is the number of observations. Thus, standard thin plate splines are not practical, as the number of observations for data mining applications is often in the millions. We have developed a finite element approximation of a thin plate spline that can handle data sizes with millions of records. Each observation record has to be read from an external file once only and there is no need to store the data in memory. The resolution of the finite element method can be chosen independently from the number of data records. An overlapping domain partitioning is applied to achieve parallelism. Our algorithm is scalable both in the number of data points as well as with the number of processors. We present first results on a Sun shared-memory multiprocessor.
dc.format.extent385105 bytes
dc.format.extent356 bytes
dc.format.mimetypeapplication/pdf
dc.format.mimetypeapplication/octet-stream
dc.language.isoen_AU
dc.subjectthin plate splines
dc.subjectfinite element method
dc.subjectparallel computing
dc.subjectlinear system
dc.subjectTE-CS
dc.titleA scalable parallel FEM surface fitting algorithm for data mining
dc.typeWorking/Technical Paper
local.description.refereedno
local.identifier.citationmonthjan
local.identifier.citationyear2001
local.identifier.eprintid1547
local.rights.ispublishedyes
dc.date.issued2001
local.contributor.affiliationDepartment of Computer Science, FEIT
local.contributor.affiliationANU
local.citationTR-CS-01-01
CollectionsANU Research Publications

Download

File Description SizeFormat Image
TR-CS-01-01.pdf376.08 kBAdobe PDFThumbnail


Items in Open Research are protected by copyright, with all rights reserved, unless otherwise indicated.

Updated:  17 November 2022/ Responsible Officer:  University Librarian/ Page Contact:  Library Systems & Web Coordinator