Contraction rates for sparse variational approximations in Gaussian process regression

Authors: Dennis Nieman, Botond Szabo, Harry van Zanten

JMLR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We study the theoretical properties of a variational Bayes method in the Gaussian Process regression model. ... The theoretical findings are demonstrated by numerical experiments. ... Finally we conclude our results with a brief numerical study in Section 7. ... 7. Numerical experiments We illustrate the theoretical results by two numerical experiments, varying both the kernel and the choice of inducing variables.
Researcher Affiliation Academia Dennis Nieman EMAIL Department of Mathematics Vrije Universiteit Amsterdam De Boelelaan 1111, 1081 HV Amsterdam The Netherlands; Botond Szabo EMAIL Department of Decision Sciences, Bocconi Institute for Data Science and Analytics, Bocconi University Via Roentgen 1, Milano, Italy; Harry van Zanten EMAIL Department of Mathematics Vrije Universiteit Amsterdam De Boelelaan 1111, 1081 HV Amsterdam The Netherlands
Pseudocode No No structured pseudocode or algorithm blocks are explicitly present in the paper. The methodology is described using mathematical formulations.
Open Source Code No The paper does not contain any explicit statements about providing open-source code, nor does it include links to code repositories.
Open Datasets No 7.1 Matérn kernel method 1 We simulate n = 3000 samples xi uniform[0, 1] and yi N(f0(xi), σ2) with σ = 0.2 and f0(x) = |x 0.4|α |x 0.2|α for α = 0.6 ... 7.2 Squared exponential kernel method 2 In a similar fashion, we simulate n = 5000 samples xi N(0, 1) and yi from the N(f0(xi), σ2) distribution with f0(x) = |x + 1|α |x + 3/2|α for α = 0.8 and σ = 0.2. The datasets are simulated for the numerical experiments, not taken from publicly available sources.
Dataset Splits No The numerical experiments use simulated data, but the context is not about model evaluation on typical train/test/validation splits. It involves simulating data to analyze posterior contraction rates. Therefore, specific dataset splits for reproduction are not applicable or mentioned.
Hardware Specification No The paper does not provide any specific details about the hardware used for the numerical experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup Yes 7. Numerical experiments We illustrate the theoretical results by two numerical experiments, varying both the kernel and the choice of inducing variables. 7.1 Matérn kernel method 1 We simulate n = 3000 samples xi uniform[0, 1] and yi N(f0(xi), σ2) with σ = 0.2 and f0(x) = |x 0.4|α |x 0.2|α for α = 0.6 ... We use the Matérn-α kernel for the GP prior and study the variational posterior using the inducing variables obtained from the covariance matrix (Section 5.1). ... 7.2 Squared exponential kernel method 2 In a similar fashion, we simulate n = 5000 samples xi N(0, 1) and yi from the N(f0(xi), σ2) distribution with f0(x) = |x + 1|α |x + 3/2|α for α = 0.8 and σ = 0.2. ... We use the squared exponential kernel as defined in (32) with b = bn = 4n 1/(1+2α) and the variational Bayes method with operator eigenvectors (Section 5.2) as the inducing variables.