Automated Variational Inference for Gaussian Process Models

Authors: Trung V Nguyen, Edwin V. Bonilla

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our approach is thoroughly verified on five models using six benchmark datasets, performing as well as the exact or hard-coded implementations while running orders of magnitude faster than the alternative MCMC sampling approaches.
Researcher Affiliation Collaboration Trung V. Nguyen ANU & NICTA Van Trung.Nguyen@nicta.com.au Edwin V. Bonilla The University of New South Wales e.bonilla@unsw.edu.au
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes We perform experiments with five GP models... on six datasets (see Table 1)... Dataset Ntrain Ntest D Likelihood p(y|f) Model Mining disasters... Boston housing... Creep... Abalone... Breast cancer... USPS...
Dataset Splits No Table 1 lists Ntrain and Ntest but no explicit validation set split percentages or counts are provided for the datasets used in the experiments. The term 'validation' in the paper refers to parameter optimization criteria, not a dataset split.
Hardware Specification Yes Training time was measured on a desktop with Intel(R) i7-2600 3.40GHz CPU with 8GB of RAM using Matlab R2012a.
Software Dependencies Yes Training time was measured on a desktop with Intel(R) i7-2600 3.40GHz CPU with 8GB of RAM using Matlab R2012a.
Experiment Setup Yes Experimental settings. The squared exponential covariance function with automatic relevance determination (see Ch. 4 in [1]) is used with the GP regression and warped GPs. The isotropic covariance is used with all other models. The noisy gradients of the ELBO are approximated with 2000 samples and 200 samples are used with control variates to reduce the variance of these estimators. The model parameters (variational, covariance hyperparameters and likelihood parameters) are learned by iteratively optimizing one set while fixing the others until convergence, which is determined when changes are less than 1e-5 for the ELBO or 1e-3 for the variational parameters.