Extended Lanczos Bidiagonalization for Dimension Reduction in Information Retrieval
نویسندگان
چکیده
We describe an extended bidiagonalization scheme designed to compute low-rank approximations of very large data matrices. Its goal is identical to that of the truncated singular value decomposition, but it is significantly cheaper. It consists in an extension of the standard Lanczos bidiagonalization that improves its approximation capabilities, while keeping the computational cost reasonable. This low-rank approximation yields much cheaper computations of the matrix-vector products that are central in many information retrieval tasks. We demonstrate effectiveness of this approach on applications in face recognition and latent semantic indexing.
منابع مشابه
Augmented Implicitly Restarted Lanczos Bidiagonalization Methods
New restarted Lanczos bidiagonalization methods for the computation of a few of the largest or smallest singular values of a large matrix are presented. Restarting is carried out by augmentation of Krylov subspaces that arise naturally in the standard Lanczos bidiagonalization method. The augmenting vectors are associated with certain Ritz or harmonic Ritz vectors. Computed examples show the ne...
متن کاملProbabilistic Bounds for the Matrix Condition Number with Extended Lanczos Bidiagonalization
Reliable estimates for the condition number of a large, sparse, real matrix A are important in many applications. To get an approximation for the condition number κ(A), an approximation for the smallest singular value is needed. Standard Krylov subspaces are usually unsuitable for finding a good approximation to the smallest singular value. Therefore, we study extended Krylov subspaces which tu...
متن کاملGreedy Tikhonov regularization for large linear ill-posed problems
Several numerical methods for the solution of large linear ill-posed problems combine Tikhonov regularization with an iterative method based on partial Lanczos bidiagonalization of the operator. This paper discusses the determination of the regularization parameter and the dimension of the Krylov subspace for this kind of methods. A method that requires a Krylov subspace of minimal dimension is...
متن کاملFiltered Matrix-Vector Products via the Lanczos Algorithm with Applications to Dimension Reduction
This paper discusses an efficient technique for computing filtered matrix-vector (mat-vec) products by exploiting the Lanczos algorithm. The goal of the proposed method, which is the same as that of the truncated singular value decomposition (SVD), is to preserve the quality of the resulting mat-vec product in major singular directions of the matrix. Unlike the SVD-based techniques, the propose...
متن کامل