Limited-Memory Matrix Methods with Applications

dc.contributor.authorKolda, Tamara G.en_US
dc.date.accessioned2004-05-31T21:07:02Z
dc.date.available2004-05-31T21:07:02Z
dc.date.created1997-04en_US
dc.date.issued1998-10-15en_US
dc.description.abstractThe focus of this dissertation is on matrix decompositions that use a limited amount of computer memory, thereby allowing problems with a very large number of variables to be solved. Specifically, we will focus on two applications areas: optimization and information retrieval. We introduce a general algebraic form for the matrix update in limited-memory quasi-Newton methods. Many well-known methods such as limited-memory Broyden Family methods satisfy the general form. We are able to prove several results about methods which satisfy the general form. In particular, we show that the only limited-memory Broyden Family method (using exact line searches) that is guaranteed to terminate within n iterations on an n-dimensional strictly convex quadratic is the limited-memory BFGS method. Furthermore, we are able to introduce several new variations on the limited-memory BFGS method that retain the quadratic termination property. We also have a new result that shows that full-memory Broyden Family methods (using exact line searches) that skip p updates to the quasi-Newton matrix will terminate in no more than n+p steps on an n-dimensional strictly convex quadratic. We propose several new variations on the limited-memory BFGS method and test these on standard test problems. We also introduce and test a new method for a process known as Latent Semantic Indexing (LSI) for information retrieval. The new method replaces the singular value matrix decomposition (SVD) at the heart of LSI with a semi-discrete matrix decomposition (SDD). We show several convergence results for the SDD and compare some strategies for computing it on general matrices. We also compare the SVD-based LSI to the SDD-based LSI and show that the SDD-based method has a faster query computation time and requires significantly less storage. We also propose and test several SDD-updating strategies for adding new documents to the collection.en_US
dc.format.extent909648 bytes
dc.format.mimetypeapplication/postscript
dc.identifier.urihttp://hdl.handle.net/1903/483
dc.language.isoen_US
dc.relation.isAvailableAtDigital Repository at the University of Marylanden_US
dc.relation.isAvailableAtUniversity of Maryland (College Park, Md.)en_US
dc.relation.isAvailableAtTech Reports in Computer Science and Engineeringen_US
dc.relation.isAvailableAtComputer Science Department Technical Reportsen_US
dc.relation.ispartofseriesUM Computer Science Department; CS-TR-3806en_US
dc.titleLimited-Memory Matrix Methods with Applicationsen_US
dc.typeTechnical Reporten_US

Files

Original bundle

Now showing 1 - 2 of 2
No Thumbnail Available
Name:
CS-TR-3806.ps
Size:
888.33 KB
Format:
Postscript Files
Loading...
Thumbnail Image
Name:
CS-TR-3806.pdf
Size:
664.88 KB
Format:
Adobe Portable Document Format
Description:
Auto-generated copy of CS-TR-3806.ps