Giga-scale Kernel Matrix-Vector Multiplication on GPU - CNRS - Centre national de la recherche scientifique Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Giga-scale Kernel Matrix-Vector Multiplication on GPU

Résumé

Kernel matrix-vector multiplication (KMVM) is a foundational operation in machine learning and scientific computing. However, as KMVM tends to scale quadratically in both memory and time, applications are often limited by these computational constraints. In this paper, we propose a novel approximation procedure coined Faster-Fast and Free Memory Method (F 3 M) to address these scaling issues of KMVM for tall (10 8 ∼ 10 9) and skinny (D ≤ 7) data. Extensive experiments demonstrate that F 3 M has empirical linear time and memory complexity with a relative error of order 10 −3 and can compute a full KMVM for a billion points in under a minute on a high-end GPU, leading to a significant speed-up in comparison to existing CPU methods. We demonstrate the utility of our procedure by applying it as a drop-in for the state-of-the-art GPU-based linear solver FALKON, improving speed 1.5-5.5 times at the cost of < 1% drop in accuracy. We further demonstrate competitive results on Gaussian Process regression coupled with significant speedups on a variety of real-world datasets.
Fichier principal
Vignette du fichier
1671_giga_scale_kernel_matrix_vecto.pdf (1.74 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03973169 , version 1 (04-02-2023)

Identifiants

Citer

Robert Hu, Siu Lun Chau, Dino Sejdinovic, Joan Alexis Glaunès. Giga-scale Kernel Matrix-Vector Multiplication on GPU. Thirty-sixth Conference on Neural Information Processing Systems, Nov 2022, New Orleans, United States. ⟨hal-03973169⟩
5 Consultations
18 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More