Documentation | Notebooks | API reference

This repository is the official implementation of the methods in the publication:

• V. Adam, P.E. Chang, M.E. Khan, and A. Solin (2021). Dual Parameterization of Sparse Variational Gaussian Processes. In Advances in Neural Information Processing Systems (NeurIPS). [arXiv]

The paper's main result shows that an alternative (dual) parameterization for SVGP models leads to a better objective for learning and allows for faster inference via natural gradient descent.

The repository has the following folder structure:

• scr contains the source code
• experiments contains scripts to reproduce some of the experiments presented in the paper
• docs contains documentation in the form of notebooks and an api reference.
• tests contains unit and integration tests for the source code

We recommend using Python version 3.7.3 and pip version 20.1.1. To install the package, run:

pip install -e .

To run the tests, notebooks, build the docs or run the experiments, install the dependencies:

pip install \
  -r tests_requirements.txt \
  -r notebook_requirements.txt \
  -r docs/docs_requirements.txt \
  -e .

To build the notebooks from source, use jupytext:

jupytext --to notebook [filename].py

If you use the code in this repository for your research, please cite the paper as follows:

@inproceedings{adam2021dual,
  title={Dual Parameterization of Sparse Variational {G}aussian Processes},
  author={Adam, Vincent and Chang, Paul Edmund and Khan, Mohammad Emtiyaz and Solin, Arno},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2021}
}

For all correspondence, please contact vincenta@gatsby.ucl.ac.uk.

This software is provided under the MIT license.