DocumentCode :
2913466
Title :
Improved Approximation Algorithms for Large Matrices via Random Projections
Author :
Sarlós, Tamás
Author_Institution :
Eotvos Univ., Budapest
fYear :
2006
fDate :
Oct. 2006
Firstpage :
143
Lastpage :
152
Abstract :
Several results appeared that show significant reduction in time for matrix multiplication, singular value decomposition as well as linear (lscr2) regression, all based on data dependent random sampling. Our key idea is that low dimensional embeddings can be used to eliminate data dependence and provide more versatile, linear time pass efficient matrix computation. Our main contribution is summarized as follows. 1) Independent of the results of Har-Peled and of Deshpande and Vempala, one of the first - and to the best of our knowledge the most efficient - relative error (1 + epsi) parA $AkparF approximation algorithms for the singular value decomposition of an m times n matrix A with M non-zero entries that requires 2 passes over the data and runs in time O((M(k/epsi+k log k) + (n+m)(k/epsi+k log k)2)log (1/sigma)). 2) The first o(nd2) time (1 + epsi) relative error approximation algorithm for n times d linear (lscr2) regression. 3) A matrix multiplication and norm approximation algorithm that easily applies to implicitly given matrices and can be used as a black box probability boosting tool
Keywords :
approximation theory; computational complexity; matrix multiplication; random processes; regression analysis; sampling methods; singular value decomposition; black box probability boosting tool; computational complexity; data dependent random sampling; large matrices; linear regression; matrix multiplication; norm approximation; random projections; relative error approximation; singular value decomposition; Algorithm design and analysis; Approximation algorithms; Automation; Boosting; Embedded computing; Linear algebra; Matrix decomposition; Sampling methods; Singular value decomposition; Sparse matrices;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Foundations of Computer Science, 2006. FOCS '06. 47th Annual IEEE Symposium on
Conference_Location :
Berkeley, CA
ISSN :
0272-5428
Print_ISBN :
0-7695-2720-5
Type :
conf
DOI :
10.1109/FOCS.2006.37
Filename :
4031351
Link To Document :
بازگشت