FMR: Fast randomized algorithms for covariance matrix computations - Archive ouverte HAL Access content directly
Poster Communications Year : 2016

FMR: Fast randomized algorithms for covariance matrix computations

Abstract

We present an open-source library implementing fast algorithms for covari-ance matrices computations, e.g., randomized low-rank approximations (LRA) and fast multipole matrix multiplication (FMM). The library can be used to approximate square roots of low-rank covariance matrices in O(N 2) operations in SVD form using randomized LRA, instead of the standard O(N 3) cost. Low-rank covariance matrices given as kernels, e.g., Gaussian decay, evaluated on 3D grids can be decomposed in O(N) operations using the FMM. The performance of the library is illustrated on two examples: • Generation of Gaussian Random Fields (GRF) on large spatial grids • MultiDimensional Scaling (MDS) for the classification of species.
Fichier principal
Vignette du fichier
poster-projet.pdf (770.67 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-01334747 , version 1 (23-06-2016)

Identifiers

  • HAL Id : hal-01334747 , version 1

Cite

Pierre Blanchard, Olivier Coulaud, Eric Darve, Alain Franc. FMR: Fast randomized algorithms for covariance matrix computations. Platform for Advanced Scientific Computing (PASC), Jun 2016, Lausanne, Switzerland. 2016. ⟨hal-01334747⟩
643 View
500 Download

Share

Gmail Mastodon Facebook X LinkedIn More