LWI-SVD: Low-rank, windowed, incremental singular value decompositions on time-evolving data sets

Xilun Chen, Kasim Candan

Research output: Chapter in Book/Report/Conference proceedingConference contribution

11 Scopus citations

Abstract

Singular Value Decomposition (SVD) is computationally costly and therefore a naive implementation does not scale to the needs of scenarios where data evolves continuously. While there are various on-line analysis and incremental decomposition techniques, these may not accurately represent the data or may be slow for the needs of many applications. To address these challenges, in this paper, we propose a Low-rank, Windowed, Incremental SVD (LWI-SVD) algorithm, which (a) leverages efficient and accurate low-rank approximations to speed up incremental SVD updates and (b) uses a window-based approach to aggregate multiple incoming updates (insertions or deletions of rows and columns) and, thus, reduces on- line processing costs. We also present an LWI-SVD with restarts (LWI2-SVD) algorithm which leverages a novel highly efficient partial reconstruction based change detection scheme to support timely refreshing of the decomposition with significant changes in the data and prevent accumulation of errors over time. Experiment results, including comparisons to other state of the art techniques on different data sets and under different parameter settings, confirm that LWI-SVD and LWI2-SVD are both efficient and accurate in maintaining decompositions.

Original languageEnglish (US)
Title of host publicationKDD 2014 - Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
PublisherAssociation for Computing Machinery
Pages987-996
Number of pages10
ISBN (Print)9781450329569
DOIs
StatePublished - Jan 1 2014
Event20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2014 - New York, NY, United States
Duration: Aug 24 2014Aug 27 2014

Publication series

NameProceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Conference

Conference20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2014
CountryUnited States
CityNew York, NY
Period8/24/148/27/14

Keywords

  • data streams
  • incremental singular value decomposition

ASJC Scopus subject areas

  • Software
  • Information Systems

Fingerprint Dive into the research topics of 'LWI-SVD: Low-rank, windowed, incremental singular value decompositions on time-evolving data sets'. Together they form a unique fingerprint.

  • Cite this

    Chen, X., & Candan, K. (2014). LWI-SVD: Low-rank, windowed, incremental singular value decompositions on time-evolving data sets. In KDD 2014 - Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 987-996). (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining). Association for Computing Machinery. https://doi.org/10.1145/2623330.2623671