# Sublinear Time Numerical Linear Algebra for Structured Matrices

## DOI:

https://doi.org/10.1609/aaai.v33i01.33014918## Abstract

We show how to solve a number of problems in numerical linear algebra, such as least squares regression, *l*_{p}-regression for any *p* ≥ 1, low rank approximation, and kernel regression, in time *T*(*A*)poly(log(*nd*)), where for a given input matrix *A* ∈ R^{n×d}, *T*(*A*) is the time needed to compute *A* · *y* for an arbitrary vector *y* ∈ R^{d}. Since *T*(*A*) ≤ *O*(nnz(*A*)), where nnz(*A*) denotes the number of non-zero entries of *A*, the time is no worse, up to polylogarithmic factors, as all of the recent advances for such problems that run in input-sparsity time. However, for many applications, *T*(*A*) can be much smaller than nnz(*A*), yielding significantly sublinear time algorithms. For example, in the overconstrained (1+*ε*)-approximate polynomial interpolation problem, *A* is a Vandermonde matrix and *T*(*A*) = *O*(*n* log *n*); in this case our running time is *n* · poly (log *n*) + poly (*d*/*ε*) and we recover the results of Avron, Sindhwani, and Woodruff (2013) as a special case. For overconstrained autoregression, which is a common problem arising in dynamical systems, *T*(*A*) = *O*(*n* log *n*), and we immediately obtain *n*· poly (log *n*) + poly(*d*/*ε*) time. For kernel autoregression, we significantly improve the running time of prior algorithms for general kernels. For the important case of autoregression with the polynomial kernel and arbitrary target vector *b* ∈ R^{n}, we obtain even faster algorithms. Our algorithms show that, perhaps surprisingly, most of these optimization problems do not require much more time than that of a polylogarithmic number of matrix-vector multiplications.

## Downloads

## Published

## How to Cite

*Proceedings of the AAAI Conference on Artificial Intelligence*,

*33*(01), 4918-4925. https://doi.org/10.1609/aaai.v33i01.33014918