Far-field compression for fast kernel summation methods in high dimensions
Abstract
We consider fast kernel summations in high dimensions: given a large set of points in d dimensions (with ) and a pair-potential function (the kernel function), we compute a weighted sum of all pairwise kernel interactions for each point in the set. Direct summation is equivalent to a (dense) matrix–vector multiplication and scales quadratically with the number of points. Fast kernel summation algorithms reduce this cost to log-linear or linear complexity. Treecodes and Fast Multipole Methods (FMMs) deliver tremendous speedups by constructing approximate representations of interactions of points that are far from each other. In algebraic terms, these representations correspond to low-rank approximations of blocks of the overall interaction matrix. Existing approaches require an excessive number of kernel evaluations with increasing d and number of points in the dataset. To address this issue, we use a randomized algebraic approach in which we first sample the rows of a block and then construct its approximate, low-rank interpolative decomposition. We examine the feasibility of this approach theoretically and experimentally. We provide a new theoretical result showing a tighter bound on the reconstruction error from uniformly sampling rows than the existing state-of-the-art. We demonstrate that our sampling approach is competitive with existing (but prohibitively expensive) methods from the literature. We also construct kernel matrices for the Laplacian, Gaussian, and polynomial kernels—all commonly used in physics and data analysis. We explore the numerical properties of blocks of these matrices, and show that they are amenable to our approach. Depending on the data set, our randomized algorithm can successfully compute low rank approximations in high dimensions. We report results for data sets with ambient dimensions from four to 1,000.
- Authors:
- Publication Date:
- Research Org.:
- Univ. of Texas, Austin, TX (United States)
- Sponsoring Org.:
- USDOE Office of Science (SC), Advanced Scientific Computing Research (ASCR)
- OSTI Identifier:
- 1804903
- Alternate Identifier(s):
- OSTI ID: 1352694; OSTI ID: 1533447
- Grant/Contract Number:
- SC0010518; SC0009286; FG02-08ER2585
- Resource Type:
- Published Article
- Journal Name:
- Applied and Computational Harmonic Analysis
- Additional Journal Information:
- Journal Name: Applied and Computational Harmonic Analysis Journal Volume: 43 Journal Issue: 1; Journal ID: ISSN 1063-5203
- Publisher:
- Elsevier
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 97 MATHEMATICS AND COMPUTING; kernel independent fast multipole methods; fast summation; randomized matrix approximation; interpolative decomposition; matrix sampling
Citation Formats
March, William B., and Biros, George. Far-field compression for fast kernel summation methods in high dimensions. United States: N. p., 2017.
Web. doi:10.1016/j.acha.2015.09.007.
March, William B., & Biros, George. Far-field compression for fast kernel summation methods in high dimensions. United States. https://doi.org/10.1016/j.acha.2015.09.007
March, William B., and Biros, George. Sat .
"Far-field compression for fast kernel summation methods in high dimensions". United States. https://doi.org/10.1016/j.acha.2015.09.007.
@article{osti_1804903,
title = {Far-field compression for fast kernel summation methods in high dimensions},
author = {March, William B. and Biros, George},
abstractNote = {We consider fast kernel summations in high dimensions: given a large set of points in d dimensions (with d>>3) and a pair-potential function (the kernel function), we compute a weighted sum of all pairwise kernel interactions for each point in the set. Direct summation is equivalent to a (dense) matrix–vector multiplication and scales quadratically with the number of points. Fast kernel summation algorithms reduce this cost to log-linear or linear complexity. Treecodes and Fast Multipole Methods (FMMs) deliver tremendous speedups by constructing approximate representations of interactions of points that are far from each other. In algebraic terms, these representations correspond to low-rank approximations of blocks of the overall interaction matrix. Existing approaches require an excessive number of kernel evaluations with increasing d and number of points in the dataset. To address this issue, we use a randomized algebraic approach in which we first sample the rows of a block and then construct its approximate, low-rank interpolative decomposition. We examine the feasibility of this approach theoretically and experimentally. We provide a new theoretical result showing a tighter bound on the reconstruction error from uniformly sampling rows than the existing state-of-the-art. We demonstrate that our sampling approach is competitive with existing (but prohibitively expensive) methods from the literature. We also construct kernel matrices for the Laplacian, Gaussian, and polynomial kernels—all commonly used in physics and data analysis. We explore the numerical properties of blocks of these matrices, and show that they are amenable to our approach. Depending on the data set, our randomized algorithm can successfully compute low rank approximations in high dimensions. We report results for data sets with ambient dimensions from four to 1,000.},
doi = {10.1016/j.acha.2015.09.007},
journal = {Applied and Computational Harmonic Analysis},
number = 1,
volume = 43,
place = {United States},
year = {Sat Jul 01 00:00:00 EDT 2017},
month = {Sat Jul 01 00:00:00 EDT 2017}
}
https://doi.org/10.1016/j.acha.2015.09.007
Web of Science
Works referenced in this record:
The black-box fast multipole method
journal, December 2009
- Fong, William; Darve, Eric
- Journal of Computational Physics, Vol. 228, Issue 23
Finding Structure with Randomness: Probabilistic Algorithms for Constructing Approximate Matrix Decompositions
journal, January 2011
- Halko, N.; Martinsson, P. G.; Tropp, J. A.
- SIAM Review, Vol. 53, Issue 2
ASKIT: Approximate Skeletonization Kernel-Independent Treecode in High Dimensions
journal, January 2015
- March, William B.; Xiao, Bo; Biros, George
- SIAM Journal on Scientific Computing, Vol. 37, Issue 2
Fast Algorithms for Polynomial Interpolation, Integration, and Differentiation
journal, October 1996
- Dutt, A.; Gu, M.; Rokhlin, V.
- SIAM Journal on Numerical Analysis, Vol. 33, Issue 5
User-Friendly Tail Bounds for Sums of Random Matrices
journal, August 2011
- Tropp, Joel A.
- Foundations of Computational Mathematics, Vol. 12, Issue 4
On the Compression of Low Rank Matrices
journal, January 2005
- Cheng, H.; Gimbutas, Z.; Martinsson, P. G.
- SIAM Journal on Scientific Computing, Vol. 26, Issue 4
The Fast Multipole Method: Numerical Implementation
journal, May 2000
- Darve, Eric
- Journal of Computational Physics, Vol. 160, Issue 1
Extensions of Lipschitz mappings into a Hilbert space
book, January 1984
- Johnson, William B.; Lindenstrauss, Joram
- Conference on Modern Analysis and Probability
Randomized Approximation of the Gram Matrix: Exact Computation and Probabilistic Bounds
journal, January 2015
- Holodnak, John T.; Ipsen, Ilse C. F.
- SIAM Journal on Matrix Analysis and Applications, Vol. 36, Issue 1
CUR matrix decompositions for improved data analysis
journal, January 2009
- Mahoney, Michael W.; Drineas, Petros
- Proceedings of the National Academy of Sciences, Vol. 106, Issue 3
A kernel-independent adaptive fast multipole algorithm in two and three dimensions
journal, May 2004
- Ying, Lexing; Biros, George; Zorin, Denis
- Journal of Computational Physics, Vol. 196, Issue 2
Fast Monte Carlo Algorithms for Matrices I: Approximating Matrix Multiplication
journal, January 2006
- Drineas, Petros; Kannan, Ravi; Mahoney, Michael W.
- SIAM Journal on Computing, Vol. 36, Issue 1
Efficient Algorithms for Computing a Strong Rank-Revealing QR Factorization
journal, July 1996
- Gu, Ming; Eisenstat, Stanley C.
- SIAM Journal on Scientific Computing, Vol. 17, Issue 4
Fast Approximation of the Discrete Gauss Transform in Higher Dimensions
journal, July 2012
- Griebel, Michael; Wissel, Daniel
- Journal of Scientific Computing, Vol. 55, Issue 1
An Implementation of the Fast Multipole Method without Multipoles
journal, July 1992
- Anderson, Christopher R.
- SIAM Journal on Scientific and Statistical Computing, Vol. 13, Issue 4
Algorithms for Numerical Analysis in High Dimensions
journal, January 2005
- Beylkin, Gregory; Mohlenkamp, Martin J.
- SIAM Journal on Scientific Computing, Vol. 26, Issue 6
An Accelerated Kernel-Independent Fast Multipole Method in One Dimension
journal, January 2007
- Martinsson, P. G.; Rokhlin, V.
- SIAM Journal on Scientific Computing, Vol. 29, Issue 3
Grid-Multipole Calculations
journal, September 1995
- Berman, C. Leonard
- SIAM Journal on Scientific Computing, Vol. 16, Issue 5
An Improved Fast Multipole Algorithm for Potential Fields
journal, November 1998
- Hrycak, Tomasz; Rokhlin, Vladimir
- SIAM Journal on Scientific Computing, Vol. 19, Issue 6
An Efficient Program for Many-Body Simulation
journal, January 1985
- Appel, Andrew W.
- SIAM Journal on Scientific and Statistical Computing, Vol. 6, Issue 1
A hierarchical O(N log N) force-calculation algorithm
journal, December 1986
- Barnes, Josh; Hut, Piet
- Nature, Vol. 324, Issue 6096
Clustering Large Graphs via the Singular Value Decomposition
journal, July 2004
- Drineas, P.; Frieze, A.; Kannan, R.
- Machine Learning, Vol. 56, Issue 1-3
Fast computation of low-rank matrix approximations
journal, April 2007
- Achlioptas, Dimitris; Mcsherry, Frank
- Journal of the ACM, Vol. 54, Issue 2
A Generalized Fast Multipole Method for Nonoscillatory Kernels
journal, January 2003
- Gimbutas, Zydrunas; Rokhlin, Vladimir
- SIAM Journal on Scientific Computing, Vol. 24, Issue 3
Computational Advertising: Techniques for Targeting Relevant Ads
journal, January 2014
- Woodruff, David P.
- Foundations and Trends® in Theoretical Computer Science, Vol. 10, Issue 1-2
Exact Matrix Completion via Convex Optimization
journal, April 2009
- Candès, Emmanuel J.; Recht, Benjamin
- Foundations of Computational Mathematics, Vol. 9, Issue 6
The Fast Gauss Transform
journal, January 1991
- Greengard, Leslie; Strain, John
- SIAM Journal on Scientific and Statistical Computing, Vol. 12, Issue 1
The Fast Multipole Method I: Error Analysis and Asymptotic Complexity
journal, January 2000
- Darve, Eric
- SIAM Journal on Numerical Analysis, Vol. 38, Issue 1
A fast randomized algorithm for the approximation of matrices
journal, November 2008
- Woolfe, Franco; Liberty, Edo; Rokhlin, Vladimir
- Applied and Computational Harmonic Analysis, Vol. 25, Issue 3
An elementary proof of a theorem of Johnson and Lindenstrauss
journal, November 2002
- Dasgupta, Sanjoy; Gupta, Anupam
- Random Structures and Algorithms, Vol. 22, Issue 1
Generalized Gaussian Quadratures and Singular Value Decompositions of Integral Operators
journal, January 1998
- Yarvin, N.; Rokhlin, V.
- SIAM Journal on Scientific Computing, Vol. 20, Issue 2
Fast Monte Carlo Algorithms for Matrices II: Computing a Low-Rank Approximation to a Matrix
journal, January 2006
- Drineas, Petros; Kannan, Ravi; Mahoney, Michael W.
- SIAM Journal on Computing, Vol. 36, Issue 1
Fast monte-carlo algorithms for finding low-rank approximations
journal, November 2004
- Frieze, Alan; Kannan, Ravi; Vempala, Santosh
- Journal of the ACM, Vol. 51, Issue 6
Improved Analysis of the Subsampled Randomized Hadamard Transform
journal, April 2011
- Tropp, Joel A.
- Advances in Adaptive Data Analysis, Vol. 03, Issue 01n02
Least Squares Support Vector Machine Classifiers
journal, June 1999
- Suykens, J. A. K.; Vandewalle, J.
- Neural Processing Letters, Vol. 9, Issue 3, p. 293-300
An Improved Fast Multipole Algorithm for Potential Fields on the Line
journal, January 1999
- Yarvin, Norman; Rokhlin, Vladimir
- SIAM Journal on Numerical Analysis, Vol. 36, Issue 2
Faster least squares approximation
journal, October 2010
- Drineas, Petros; Mahoney, Michael W.; Muthukrishnan, S.
- Numerische Mathematik, Vol. 117, Issue 2
A fast algorithm for particle simulations
journal, December 1987
- Greengard, L.; Rokhlin, V.
- Journal of Computational Physics, Vol. 73, Issue 2
Improved Matrix Algorithms via the Subsampled Randomized Hadamard Transform
journal, January 2013
- Boutsidis, Christos; Gittens, Alex
- SIAM Journal on Matrix Analysis and Applications, Vol. 34, Issue 3
Sampling from large matrices: An approach through geometric functional analysis
journal, July 2007
- Rudelson, Mark; Vershynin, Roman
- Journal of the ACM, Vol. 54, Issue 4
A randomized algorithm for the decomposition of matrices
journal, January 2011
- Martinsson, Per-Gunnar; Rokhlin, Vladimir; Tygert, Mark
- Applied and Computational Harmonic Analysis, Vol. 30, Issue 1