Log In Sign Up

Krylov Subspace Recycling for Fast Iterative Least-Squares in Machine Learning

by   Filip de Roos, et al.

Solving symmetric positive definite linear problems is a fundamental computational task in machine learning. The exact solution, famously, is cubicly expensive in the size of the matrix. To alleviate this problem, several linear-time approximations, such as spectral and inducing-point methods, have been suggested and are now in wide use. These are low-rank approximations that choose the low-rank space a priori and do not refine it over time. While this allows linear cost in the data-set size, it also causes a finite, uncorrected approximation error. Authors from numerical linear algebra have explored ways to iteratively refine such low-rank approximations, at a cost of a small number of matrix-vector multiplications. This idea is particularly interesting in the many situations in machine learning where one has to solve a sequence of related symmetric positive definite linear problems. From the machine learning perspective, such deflation methods can be interpreted as transfer learning of a low-rank approximation across a time-series of numerical tasks. We study the use of such methods for our field. Our empirical results show that, on regression and classification problems of intermediate size, this approach can interpolate between low computational cost and numerical precision.


page 1

page 2

page 3

page 4


Two-level Nyström–Schur preconditioner for sparse symmetric positive definite matrices

Randomized methods are becoming increasingly popular in numerical linear...

Low-rank signal subspace: parameterization, projection and signal estimation

The paper contains several theoretical results related to the weighted n...

Randomized Nyström Preconditioning

This paper introduces the Nyström PCG algorithm for solving a symmetric ...

Spectral Algorithm for Low-rank Multitask Regression

Multitask learning, i.e. taking advantage of the relatedness of individu...

Linear Time Sinkhorn Divergences using Positive Features

Although Sinkhorn divergences are now routinely used in data sciences to...

Code Repositories


Kyrlov Subspace Recycling for Fast Iterative Least Squares in Machine Learning using Data Sparse Matrix Computations

view repo