1. ScholarBank@NUS
  2. 1. Staff
  3. Staff Publications
Please use this identifier to cite or link to this item: https://doi.org/10.1109/SSP.2018.8450822
DC FieldValue
dc.titleMinimax Lower Bounds for Nonnegative Matrix Factorization
dc.contributor.authorAlsan, M
dc.contributor.authorLiu, Z
dc.contributor.authorTan, VYF
dc.date.accessioned2020-08-31T06:39:39Z
dc.date.available2020-08-31T06:39:39Z
dc.date.issued2018-08-29
dc.identifier.citationAlsan, M, Liu, Z, Tan, VYF (2018-08-29). Minimax Lower Bounds for Nonnegative Matrix Factorization. 2018 IEEE Statistical Signal Processing Workshop (SSP) : 328-332. ScholarBank@NUS Repository. https://doi.org/10.1109/SSP.2018.8450822
dc.identifier.isbn9781538615706
dc.identifier.urihttps://scholarbank.nus.edu.sg/handle/10635/173662
dc.description.abstract© 2018 IEEE. The non-negative matrix factorization (NMF) problem consists in modeling data samples as non-negative linear combinations of non-negative dictionary vectors. While many algorithms for NMF have been proposed, fundamental performance limits of these algorithms are currently not available. This paper plugs this gap by providing lower bounds on the minimax risk (the minimum achievable worst case mean squared error) of estimating the non-negative dictionary matrix under a set of locality and statistical assumptions.
dc.publisherIEEE
dc.sourceElements
dc.typeConference Paper
dc.date.updated2020-07-22T19:31:57Z
dc.contributor.departmentELECTRICAL AND COMPUTER ENGINEERING
dc.description.doi10.1109/SSP.2018.8450822
dc.description.sourcetitle2018 IEEE Statistical Signal Processing Workshop (SSP)
dc.description.page328-332
dc.published.statePublished
dc.description.redepositcompleted
Appears in Collections:Elements
Staff Publications

Show simple item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
MINIMAX LOWER BOUNDS FOR NONNEGATIVE MATRIX FACTORIZATION.pdf554.61 kBAdobe PDF

CLOSED

None
1570431479.pdf241.63 kBAdobe PDF

CLOSED

None

Google ScholarTM

Check

Altmetric


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