A generalization of the randomized singular value decomposition

Authors: Nicolas Boulle, Alex Townsend

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical examples on matrices and HS operators demonstrate the applicability of the algorithm.
Researcher Affiliation Academia Nicolas Boull e Mathematical Institute University of Oxford Oxford, OX2 6GG, UK boulle@maths.ox.ac.uk Alex Townsend Department of Mathematics Cornell University Ithaca, NY 14853, USA townsend@cornell.edu
Pseudocode Yes Algorithm 1 Randomized SVD for HS operators
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets No The paper uses synthetic problem setups (e.g., discretized matrix of a differential operator, defined kernels) rather than publicly available datasets with concrete access information.
Dataset Splits No The paper does not provide specific dataset split information (e.g., train/validation/test percentages or counts) needed to reproduce the data partitioning.
Hardware Specification Yes Timings were performed on an Intel Xeon CPU E5-2667 v2 @ 3.30GHz using MATLAB R2020b without explicit parallelization.
Software Dependencies Yes Timings were performed on an Intel Xeon CPU E5-2667 v2 @ 3.30GHz using MATLAB R2020b without explicit parallelization. The algorithm is implemented in the Chebfun software system (Driscoll et al., 2014), which is a MATLAB package for computing with functions. The continuous Cholesky factorization is implemented in Chebfun2 (Townsend & Trefethen, 2013), which is the extension of Chebfun for computing with two-dimensional functions.
Experiment Setup Yes We vary the number of columns (i.e. samples from the GP) in the input matrix Ωfrom 1 to 2000. We employ the squared-exponential covariance kernel KSE with parameter ℓ= 0.01 and k = 100 samples (see Equation (3)) to sample random functions from the associated GP. [...] K(2,2) Jac with eigenvalues λj = 1/j3, for j ≥ 1.