Using machine learning to improve dense and sparse matrix multiplication kernels
Date
Authors
Major Professor
Advisor
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Authors
Research Projects
Organizational Units
Journal Issue
Is Version Of
Versions
Series
Department
Abstract
This work is comprised of two different projects in numerical linear algebra. The first project is about using machine learning to speed up dense matrix-matrix multiplication computations on a shared-memory computer architecture. We found that found basic loop-based matrix-matrix multiplication algorithms tied to a decision tree algorithm selector were competitive to using Intel's Math Kernel Library for the same computation. The second project is a preliminary report about re-implementing an encoding format for spare matrix-vector multiplication called Compressed Spare eXtended (CSX). The goal for the second project is to use machine learning to aid in encoding matrix substructures in the CSX format without using exhaustive search and a Just-In-Time compiler.