Meta-Learning for Stochastic Gradient MCMC

This repo contains the implementation of the experiments in the following paper:

Wenbo Gong*, Yingzhen Li* and Jose Miguel Hernandez Lobato

Meta-Learning for Stochastic Gradient MCMC

International Conference on Learning Representations (ICLR), 2019

Contributions: Wenbo and Yingzhen initiated the project together. Wenbo designed thewith architecture of the sampler, and implemented all the experiments. Yingzhen is mainly responsible for loss function design, experiment design, and the paper writing. Miguel is here because he is Wenbo's PhD supervisor, and he provided some comments on the draft.

If you use our code in your research, please consider citing the paper.

Gaussian experiments

See README in Toy Example/

MNIST experiments

See README in BNN_MNIST/

Citing the paper (bib)

@inproceedings{
gong2018metalearning,
title={Meta-Learning For Stochastic Gradient {MCMC}},
author={Wenbo Gong and Yingzhen Li and José Miguel Hernández-Lobato},
booktitle={International Conference on Learning Representations},
year={2019},
url={https://openreview.net/forum?id=HkeoOo09YX},
}