Naslov
Analyzing different versions of randomized (G)SVD for regularization of large scale discrete inverse problems

Autori
Carević, Anita ; Almekkawy, Mohamed

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Book of Abstracts - 11th Conference on Applied Mathematics and Scientific Computing / - , 2022, 22-22

Skup
11th Conference on Applied Mathematics and Scientific Computing

Mjesto i datum
Brijuni, Hrvatska, 05.09.2022. - 09.09.2022

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
Ill-posed inverse problem ; regularization ; randomized GSVD

Sažetak
Singular value decomposition (SVD) and generalized singular value decomposition (GSVD) can have an important role when regularizing an ill-posed discrete inverse problem. Since the calculation of (G)SVD for large-scale problems is time consuming, methods that can be used to accelerate it are of particular interest. Recently, research using randomized algorithms to speed up the computation of (G)SVD have been widely reported. Accordingly, existing regularization methods have been modified to use randomized (G)SVD. However, their efficiency may be questionable when regularizing an inverse problem in which the singular values of the operator matrix decay gradually without a significant gap. This is very common in application, usually when the problem arise from discretization of a Fredholm integral equation. In this talk we analyze the application of different versions of randomized (G)SVD to inverse problem in which the singular values decay gradually without a significant gap and offer some improvements for their efficiency.

Izvorni jezik
Engleski

Znanstvena područja
Matematika, Računarstvo

POVEZANOST RADA