Improving medium-grain partitioning for scalable sparse tensor decomposition

Simple item page
buir.contributor.authorAykanat, Cevdet
dc.citation.epage2825en_US
dc.citation.issueNumber12en_US
dc.citation.spage2814en_US
dc.citation.volumeNumber29en_US
dc.contributor.authorAcer, S.en_US
dc.contributor.authorTorun, T.en_US
dc.contributor.authorAykanat, Cevdeten_US
dc.date.accessioned2019-02-21T16:05:54Z
dc.date.available2019-02-21T16:05:54Z
dc.date.issued2018en_US
dc.departmentDepartment of Computer Engineeringen_US
dc.description.abstractTensor decomposition is widely used in the analysis of multi-dimensional data. The canonical polyadic decomposition (CPD) is one of the most popular decomposition methods and commonly found by the CPD-ALS algorithm. High computational and memory costs of CPD-ALS necessitate the use of a distributed-memory-parallel algorithm for efficiency. The medium-grain CPD-ALS algorithm, which adopts multi-dimensional cartesian tensor partitioning, is one of the most successful distributed CPD-ALS algorithms for sparse tensors. This is because cartesian partitioning imposes nice upper bounds on communication overheads. However, this model does not utilize the sparsity pattern of the tensor to reduce the total communication volume. The objective of this work is to fill this literature gap. We propose a novel hypergraph-partitioning model, CartHP, whose partitioning objective correctly encapsulates the minimization of total communication volume of multi-dimensional cartesian tensor partitioning. Experiments on twelve real-world tensors using up to 1024 processors validate the effectiveness of the proposed CartHP model. Compared to the baseline medium-grain model, CartHP achieves average reductions of 52, 43 and 24 percent in total communication volume, communication time and overall runtime of CPD-ALS, respectively.
dc.description.provenanceMade available in DSpace on 2019-02-21T16:05:54Z (GMT). No. of bitstreams: 1 Bilkent-research-paper.pdf: 222869 bytes, checksum: 842af2b9bd649e7f548593affdbafbb3 (MD5) Previous issue date: 2018en
dc.description.sponsorshipThis work is supported by the Scientific and Technological Research Council of Turkey (TUBITAK) under project EEEAG-116E043. We acknowledge PRACE for awarding us access to resource Hazel Hen (Cray XC40) based in Germany at HLRS.
dc.identifier.doi10.1109/TPDS.2018.2841843
dc.identifier.issn1045-9219
dc.identifier.urihttp://hdl.handle.net/11693/50280
dc.language.isoEnglish
dc.publisherInstitute of Electrical and Electronics Engineers
dc.relation.isversionofhttps://doi.org/10.1109/TPDS.2018.2841843
dc.relation.projectPartnership for Advanced Computing in Europe AISBL, PRACE - EEEAG-116E043 - Türkiye Bilimsel ve Teknolojik Araştirma Kurumu, TÜBITAK
dc.source.titleIEEE Transactions on Parallel and Distributed Systemsen_US
dc.subjectCanonical polyadic decompositionen_US
dc.subjectCartesian partitioningen_US
dc.subjectCommunication volumeen_US
dc.subjectHypergraph partitioningen_US
dc.subjectLoad balancingen_US
dc.subjectSparse tensoren_US
dc.titleImproving medium-grain partitioning for scalable sparse tensor decompositionen_US
dc.typeArticleen_US

Files

Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
Improving_Medium_Grain_Partitioning_for.pdf
Size:
804.17 KB
Format:
Adobe Portable Document Format
Description:
Full printable version
Download

Collections

Department of Computer Engineering