Sparse Gaussian Processes with Spherical Harmonic Features

Authors: Vincent Dutordoir, Nicolas Durrande, James Hensman

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments show that our model is able to fit a regression model for a dataset with 6 million entries two orders of magnitude faster compared to standard sparse GPs, while retaining state of the art accuracy. We also demonstrate competitive performance on classification with non-conjugate likelihoods. ... Section 4 is dedicated to the experimental evaluation.
Researcher Affiliation Collaboration 1PROWLER.io, Cambridge, United Kingdom 2Department of Engineering, University of Cambridge, Cambridge, United Kingdom 3Amazon Research, Cambridge, United Kingdom (work done while JH was affiliated to PROWLER.io).
Pseudocode No The paper does not contain any sections or figures explicitly labeled as 'Pseudocode' or 'Algorithm', nor any structured algorithm blocks.
Open Source Code No The paper does not include an unambiguous statement that the authors are releasing the source code for the work described, nor a direct link to a repository containing their implementation code.
Open Datasets Yes We use five UCI regression datasets to compare the performance of our method against other GP approaches. ... We use the 2008 U.S. airline delay dataset to asses these capabilities.
Dataset Splits Yes For each dataset we randomly select 90% of the data for training and 10% for testing and repeat this 5 times to get error bars. ... Every split is repeated 10 times and we report the mean and one standard deviation of the MSE and NLPD.
Hardware Specification Yes All these experiments were ran on a single consumer-grade GPU (Nvidia GTX 1070).
Software Dependencies No The paper mentions optimizers like L-BFGS and Adam, but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes For VISH we normalize the inputs so that each column falls within [ vd, vd]. ... For SVGP and VISH we first used a subset of 20,000 points to train the variational and hyper-parameters of the model with L-BFGS. We then applied Adam to the whole dataset.