Matches in SemOpenAlex for { <https://semopenalex.org/work/W3022184436> ?p ?o ?g. }
Showing items 1 to 75 of
75
with 100 items per page.
- W3022184436 abstract "Author(s): Wu, Huiwen | Advisor(s): Chen, Long | Abstract: We construct a preconditioner for solving the linear least square problems, which are simplest and most popular arising in data fitting, imaging processing and high dimension data analysis. The existed methods for solving least squares problems either has a large computational cost or depends highly on the condition number of the matrix. Recently, there is a surge of interest in developing randomized algorithms for solving least squares problems for the purpose of efficiency and scalability. We construct a new preconditioner equipped with sampling procedure to reduce computational complexity and apply Gauss Seidel iterations to grab the high frequency component of the solution, which reduces the dependence of performance of the conditioner number. Experimental studies compared with Conjugate Gradient Descent method (CG) are presented on six different simulations including dense Gaussian matrix, `semi Gaussian' matrix, Sparse random matrix, `UDV' matrix and Graph Laplacian matrix and a non-negative constrained problem to show the effectiveness of the proposed preconditioner. A general scheme for solving non-constraint convex and smooth minimization problem $ min_{x in mathcal V} f(x)$ is developed in this thesis. The scheme does gradient descent on each subspace based on a stable space decomposition of $mathcal V$. With assumptions of Lipschitz continuous of the gradient on $mathcal V$ and its subspaces, convexity or strong convexity on $mathcal V$, we prove linear convergence for strongly convex objective function for both non-uniform sampling and uniform sampling. For non-uniform sampling, the convergence depends on the expected condition number, and for uniform sampling, the convergence depends on the supreme condition number. Moreover, we also show sublinear convergence for convex function. Numerical examples on Nestrov's worst function and linear regression both outperform randomized coordinate method. We conclude that our scheme generalizes the gradient descent methods, randomized (block) coordinate descent methods and full approximation scheme." @default.
- W3022184436 created "2020-05-13" @default.
- W3022184436 creator A5057564152 @default.
- W3022184436 date "2019-01-01" @default.
- W3022184436 modified "2023-09-24" @default.
- W3022184436 title "Randomized Fast Solvers for Linear and Nonlinear Problems in Data Science" @default.
- W3022184436 hasPublicationYear "2019" @default.
- W3022184436 type Work @default.
- W3022184436 sameAs 3022184436 @default.
- W3022184436 citedByCount "0" @default.
- W3022184436 crossrefType "journal-article" @default.
- W3022184436 hasAuthorship W3022184436A5057564152 @default.
- W3022184436 hasConcept C106487976 @default.
- W3022184436 hasConcept C11413529 @default.
- W3022184436 hasConcept C119857082 @default.
- W3022184436 hasConcept C121332964 @default.
- W3022184436 hasConcept C126255220 @default.
- W3022184436 hasConcept C153258448 @default.
- W3022184436 hasConcept C158693339 @default.
- W3022184436 hasConcept C159694833 @default.
- W3022184436 hasConcept C159985019 @default.
- W3022184436 hasConcept C167431342 @default.
- W3022184436 hasConcept C192562407 @default.
- W3022184436 hasConcept C28826006 @default.
- W3022184436 hasConcept C33923547 @default.
- W3022184436 hasConcept C41008148 @default.
- W3022184436 hasConcept C50644808 @default.
- W3022184436 hasConcept C62520636 @default.
- W3022184436 hasConcept C81184566 @default.
- W3022184436 hasConcept C84545080 @default.
- W3022184436 hasConceptScore W3022184436C106487976 @default.
- W3022184436 hasConceptScore W3022184436C11413529 @default.
- W3022184436 hasConceptScore W3022184436C119857082 @default.
- W3022184436 hasConceptScore W3022184436C121332964 @default.
- W3022184436 hasConceptScore W3022184436C126255220 @default.
- W3022184436 hasConceptScore W3022184436C153258448 @default.
- W3022184436 hasConceptScore W3022184436C158693339 @default.
- W3022184436 hasConceptScore W3022184436C159694833 @default.
- W3022184436 hasConceptScore W3022184436C159985019 @default.
- W3022184436 hasConceptScore W3022184436C167431342 @default.
- W3022184436 hasConceptScore W3022184436C192562407 @default.
- W3022184436 hasConceptScore W3022184436C28826006 @default.
- W3022184436 hasConceptScore W3022184436C33923547 @default.
- W3022184436 hasConceptScore W3022184436C41008148 @default.
- W3022184436 hasConceptScore W3022184436C50644808 @default.
- W3022184436 hasConceptScore W3022184436C62520636 @default.
- W3022184436 hasConceptScore W3022184436C81184566 @default.
- W3022184436 hasConceptScore W3022184436C84545080 @default.
- W3022184436 hasLocation W30221844361 @default.
- W3022184436 hasOpenAccess W3022184436 @default.
- W3022184436 hasPrimaryLocation W30221844361 @default.
- W3022184436 hasRelatedWork W1539062885 @default.
- W3022184436 hasRelatedWork W1600618388 @default.
- W3022184436 hasRelatedWork W1978855878 @default.
- W3022184436 hasRelatedWork W1995823823 @default.
- W3022184436 hasRelatedWork W2014421268 @default.
- W3022184436 hasRelatedWork W2048305372 @default.
- W3022184436 hasRelatedWork W2181983165 @default.
- W3022184436 hasRelatedWork W2530522696 @default.
- W3022184436 hasRelatedWork W2793857849 @default.
- W3022184436 hasRelatedWork W2896361210 @default.
- W3022184436 hasRelatedWork W2940073586 @default.
- W3022184436 hasRelatedWork W2963078637 @default.
- W3022184436 hasRelatedWork W2967203827 @default.
- W3022184436 hasRelatedWork W2988331362 @default.
- W3022184436 hasRelatedWork W3105227640 @default.
- W3022184436 hasRelatedWork W3155255129 @default.
- W3022184436 hasRelatedWork W3162680648 @default.
- W3022184436 hasRelatedWork W3194355613 @default.
- W3022184436 hasRelatedWork W3205024572 @default.
- W3022184436 hasRelatedWork W2153619784 @default.
- W3022184436 isParatext "false" @default.
- W3022184436 isRetracted "false" @default.
- W3022184436 magId "3022184436" @default.
- W3022184436 workType "article" @default.