Scaling Gaussian Process Regression with Derivatives

Authors: David Eriksson, Kun Dong, Eric Lee, David Bindel, Andrew G. Wilson

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments use the squared exponential (SE) kernel... We show the relative fitting accuracy of SE, SKI, D-SE, and D-SKI on some standard test functions in Table 1. We apply active subspace pre-processing to the 20 dimensional Welsh test function in [2]. Rough terrain reconstruction is a key application in robotics [9, 15]... In the following experiment, we consider roughly 23 million non-uniformly sampled elevation measurements of Mount St. Helens obtained via Li DAR [3].
Researcher Affiliation Academia David Eriksson Center for Applied Mathematics Cornell University Ithaca, NY 14853 dme65@cornell.edu Kun Dong Center for Applied Mathematics Cornell University Ithaca, NY 14853 kd383@cornell.edu Eric Hans Lee Department of Computer Science Cornell University Ithaca, NY 14853 ehl59@cornell.edu David Bindel Department of Computer Science Cornell University Ithaca, NY 14853 bindel@cornell.edu Andrew Gordon Wilson School of Operations Research and Information Engineering Cornell University Ithaca, NY 14853 andrew@cornell.edu
Pseudocode Yes Algorithm 1: BO with derivatives and active subspace learning
Open Source Code Yes Code, experiments, and figures may be reproduced at: https://github.com/ericlee0803/GP_Derivatives.
Open Datasets Yes We train the SE kernel on 4000 points, the D-SE kernel on 4000/(d + 1) points, and SKI and D-SKI with SE kernel on 10000 points to achieve comparable runtimes between methods. In the following experiment, we consider roughly 23 million non-uniformly sampled elevation measurements of Mount St. Helens obtained via Li DAR [3]. [24] S. Surjanovic and D. Bingham. Virtual library of simulation experiments: Test functions and datasets. http://www.sfu.ca/ ssurjano, 2018. [3] Puget Sound Li DAR Consortium. Mount Saint Helens Li DAR data. University of Washington, 2002.
Dataset Splits No The paper mentions training and testing splits (e.g., 'We randomly select 90% of the grid for training and the remainder for testing.') but does not explicitly describe a separate validation set or its proportion.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running experiments.
Software Dependencies No The paper mentions various software components and methods (e.g., 'FFTs', 'cubic convolutional interpolation', 'quintic interpolation', 'conjugate gradients', 'pivoted Cholesky') but does not specify their version numbers or other software dependencies with versions.
Experiment Setup No The paper provides some experimental settings, such as number of training points (e.g., '5000 training points', '25000 points') and fixed dimensions ('d = 2'), but it lacks comprehensive details on hyperparameters, optimizer settings, or other system-level training configurations to fully reproduce the experimental setup.