Improved Algorithms for Kernel Matrix-Vector Multiplication Under Sparsity Assumptions

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Indyk, Piotr, Kapralov, Michael, Sheth, Kshiteej, Wagner, Tal
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909713854103552
author Indyk, Piotr
Kapralov, Michael
Sheth, Kshiteej
Wagner, Tal
author_facet Indyk, Piotr
Kapralov, Michael
Sheth, Kshiteej
Wagner, Tal
contents Motivated by the problem of fast processing of attention matrices, we study fast algorithms for computing matrix-vector products for asymmetric Gaussian Kernel matrices $K\in \mathbb{R}^{n\times n}$. $K$'s columns are indexed by a set of $n$ keys $k_1,k_2\ldots, k_n\in \mathbb{R}^d$, rows by a set of $n$ queries $q_1,q_2,\ldots,q_n\in \mathbb{R}^d $, and its $i,j$ entry is $K_{ij} = e^{-\|q_i-k_j\|_2^2/2σ^2}$ for some bandwidth parameter $σ>0$. Given a vector $x\in \mathbb{R}^n$ and error parameter $ε>0$, our task is to output a $y\in \mathbb{R}^n$ such that $\|Kx-y\|_2\leq ε\|x\|_2$ in time subquadratic in $n$ and linear in $d$. Our algorithms rely on the following modelling assumption about the matrices $K$: the sum of the entries of $K$ scales linearly in $n$, as opposed to worst case quadratic growth. We validate this assumption experimentally, for Gaussian kernel matrices encountered in various settings such as fast attention computation in LLMs. We obtain the first subquadratic-time algorithm that works under this assumption, for unrestricted vectors.
format Preprint
id arxiv_https___arxiv_org_abs_2507_23539
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Improved Algorithms for Kernel Matrix-Vector Multiplication Under Sparsity Assumptions
Indyk, Piotr
Kapralov, Michael
Sheth, Kshiteej
Wagner, Tal
Machine Learning
Data Structures and Algorithms
Motivated by the problem of fast processing of attention matrices, we study fast algorithms for computing matrix-vector products for asymmetric Gaussian Kernel matrices $K\in \mathbb{R}^{n\times n}$. $K$'s columns are indexed by a set of $n$ keys $k_1,k_2\ldots, k_n\in \mathbb{R}^d$, rows by a set of $n$ queries $q_1,q_2,\ldots,q_n\in \mathbb{R}^d $, and its $i,j$ entry is $K_{ij} = e^{-\|q_i-k_j\|_2^2/2σ^2}$ for some bandwidth parameter $σ>0$. Given a vector $x\in \mathbb{R}^n$ and error parameter $ε>0$, our task is to output a $y\in \mathbb{R}^n$ such that $\|Kx-y\|_2\leq ε\|x\|_2$ in time subquadratic in $n$ and linear in $d$. Our algorithms rely on the following modelling assumption about the matrices $K$: the sum of the entries of $K$ scales linearly in $n$, as opposed to worst case quadratic growth. We validate this assumption experimentally, for Gaussian kernel matrices encountered in various settings such as fast attention computation in LLMs. We obtain the first subquadratic-time algorithm that works under this assumption, for unrestricted vectors.
title Improved Algorithms for Kernel Matrix-Vector Multiplication Under Sparsity Assumptions
topic Machine Learning
Data Structures and Algorithms
url https://arxiv.org/abs/2507.23539