Minimizing Communication for Parallel Symmetric Tensor Times Same Vector Computation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Daas, Hussam Al, Ballard, Grey, Grigori, Laura, Kumar, Suraj, Rouse, Kathryn, Vérité, Mathieu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909652533379072
author Daas, Hussam Al
Ballard, Grey
Grigori, Laura
Kumar, Suraj
Rouse, Kathryn
Vérité, Mathieu
author_facet Daas, Hussam Al
Ballard, Grey
Grigori, Laura
Kumar, Suraj
Rouse, Kathryn
Vérité, Mathieu
contents In this article, we focus on the parallel communication cost of multiplying the same vector along two modes of a $3$-dimensional symmetric tensor. This is a key computation in the higher-order power method for determining eigenpairs of a $3$-dimensional symmetric tensor and in gradient-based methods for computing a symmetric CP decomposition. We establish communication lower bounds that determine how much data movement is required to perform the specified computation in parallel. The core idea of the proof relies on extending a key geometric inequality for $3$-dimensional symmetric computations. We demonstrate that the communication lower bounds are tight by presenting an optimal algorithm where the data distribution is a natural extension of the triangle block partition scheme for symmetric matrices to 3-dimensional symmetric tensors.
format Preprint
id arxiv_https___arxiv_org_abs_2506_15488
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Minimizing Communication for Parallel Symmetric Tensor Times Same Vector Computation
Daas, Hussam Al
Ballard, Grey
Grigori, Laura
Kumar, Suraj
Rouse, Kathryn
Vérité, Mathieu
Distributed, Parallel, and Cluster Computing
In this article, we focus on the parallel communication cost of multiplying the same vector along two modes of a $3$-dimensional symmetric tensor. This is a key computation in the higher-order power method for determining eigenpairs of a $3$-dimensional symmetric tensor and in gradient-based methods for computing a symmetric CP decomposition. We establish communication lower bounds that determine how much data movement is required to perform the specified computation in parallel. The core idea of the proof relies on extending a key geometric inequality for $3$-dimensional symmetric computations. We demonstrate that the communication lower bounds are tight by presenting an optimal algorithm where the data distribution is a natural extension of the triangle block partition scheme for symmetric matrices to 3-dimensional symmetric tensors.
title Minimizing Communication for Parallel Symmetric Tensor Times Same Vector Computation
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2506.15488