Hierarchical Randomized Low-Rank Approximations: Applications to covariance kernel matrices and generation of Gaussian Random Fields

Abstract : We propose a new efficient algorithm for performing hierarchical kernel MVPs in O(N) operations called the Uniform FMM (UFMM), an FFT accelerated variant of the black-box FMM by Fong and Darve. The UFMM is used to speed-up randomized low-rank methods thus reducing their computational cost to O(N) in time and memory. Numerical benchmarks include low-rank approximations of covariance matrices for the simulation of stationary random fields on very large distributions of points.
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01255724
Contributor : Pierre Blanchard <>
Submitted on : Friday, January 15, 2016 - 12:05:12 PM
Last modification on : Wednesday, April 4, 2018 - 1:24:20 AM

Identifiers

  • HAL Id : hal-01255724, version 1

Citation

Pierre Blanchard, Olivier Coulaud, E Darve, B Bramas. Hierarchical Randomized Low-Rank Approximations: Applications to covariance kernel matrices and generation of Gaussian Random Fields. SIAM Conference on Applied Linear Algebra (SIAM LA), SIAM, Oct 2015, Atlanta, United States. ⟨hal-01255724⟩

Share

Metrics

Record views

596

Files downloads

129