Abstract:This work is concerned with computing low-rank approximations of a matrix function $f(A)$ for a large symmetric positive semi-definite matrix $A$, a task that arises in, e.g., statistical learning and inverse problems. The application of popular randomized methods, such as the randomized singular value decomposition or the Nyström approximation, to $f(A)$ requires multiplying $f(A)$ with a few random vectors. A significant disadvantage of such an approach, matrix-vector products with $f(A)$ are considerably more expensive than matrix-vector products with $A$, even when carried out only approximately via, e.g., the Lanczos method. In this work, we present and analyze funNyström, a simple and inexpensive method that constructs a low-rank approximation of $f(A)$ directly from a Nyström approximation of $A$, completely bypassing the need for matrix-vector products with $f(A)$. It is sensible to use funNyström whenever $f$ is monotone and satisfies $f(0) = 0$. Under the stronger assumption that $f$ is operator monotone, which includes the matrix square root $A^{1/2}$ and the matrix logarithm $\log(I+A)$, we derive probabilistic bounds for the error in the Frobenius, nuclear, and operator norms. These bounds confirm the numerical observation that funNyström tends to return an approximation that compares well with the best low-rank approximation of $f(A)$. Furthermore, compared to existing methods, funNyström requires significantly fewer matrix-vector products with $A$ to obtain a low-rank approximation of $f(A)$, without sacrificing accuracy or reliability. Our method is also of interest when estimating quantities associated with $f(A)$, such as the trace or the diagonal entries of $f(A)$. In particular, we propose and analyze funNyström++, a combination of funNyström with the recently developed Hutch++ method for trace estimation.

Efficient direct position determination using Frobenius norm approximation

MUSIC With Capped Frobenius Norm: Efficient Robust Direction-of-Arrival Estimator

Fast and Accurate Matrix Completion Via Truncated Nuclear Norm Regularization

A direct method to Frobenius norm-based matrix regression

Fast Spectral Low Rank Matrix Approximation

Efficient Randomized Algorithms for the Fixed-Precision Low-Rank Matrix Approximation

Matrix Nearness Problems with Off-Block-Diagonal Rank Constraints

Low-Rank Approximation via Generalized Reweighted Iterative Nuclear and Frobenius Norms

Fast and Adaptive Sparse Precision Matrix Estimation in High Dimensions

A Theoretical Comparison of Two Classes of Projection Methods

Fast gridless direction-of-arrival estimation for wideband linear frequency modulated signals Based on fractional Fourier transform

Randomized low-rank approximation of monotone matrix functions

The Nearest Graph Laplacian in Frobenius Norm

Entrywise tensor-train approximation of large tensors via random embeddings

Towards More Efficient SPSD Matrix Approximation and CUR Matrix Decomposition

Algorithm-agnostic low-rank approximation of operator monotone matrix functions

A general error analysis for randomized low-rank approximation with application to data assimilation

Fast vision-based pose estimation iterative algorithm

Low Rank Approximation of Dual Complex Matrices

A Fast Gridless Covariance Matrix Reconstruction Method for One- and Two-Dimensional Direction-of-Arrival Estimation

An Efficient Algorithm for the 𝓁p Norm Based Metric Nearness Problem