Minibatch Gibbs Sampling on Large Graphical Models

Chris De Sa, Vincent Chen, Wing Wong
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:1165-1173, 2018.

Abstract

Gibbs sampling is the de facto Markov chain Monte Carlo method used for inference and learning on large scale graphical models. For complicated factor graphs with lots of factors, the performance of Gibbs sampling can be limited by the computational cost of executing a single update step of the Markov chain. This cost is proportional to the degree of the graph, the number of factors adjacent to each variable. In this paper, we show how this cost can be reduced by using minibatching: subsampling the factors to form an estimate of their sum. We introduce several minibatched variants of Gibbs, show that they can be made unbiased, prove bounds on their convergence rates, and show that under some conditions they can result in asymptotic single-update-run-time speedups over plain Gibbs sampling.

Cite this Paper


BibTeX
@InProceedings{pmlr-v80-desa18a, title = {Minibatch Gibbs Sampling on Large Graphical Models}, author = {De Sa, Chris and Chen, Vincent and Wong, Wing}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {1165--1173}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/desa18a/desa18a.pdf}, url = {https://proceedings.mlr.press/v80/desa18a.html}, abstract = {Gibbs sampling is the de facto Markov chain Monte Carlo method used for inference and learning on large scale graphical models. For complicated factor graphs with lots of factors, the performance of Gibbs sampling can be limited by the computational cost of executing a single update step of the Markov chain. This cost is proportional to the degree of the graph, the number of factors adjacent to each variable. In this paper, we show how this cost can be reduced by using minibatching: subsampling the factors to form an estimate of their sum. We introduce several minibatched variants of Gibbs, show that they can be made unbiased, prove bounds on their convergence rates, and show that under some conditions they can result in asymptotic single-update-run-time speedups over plain Gibbs sampling.} }
Endnote
%0 Conference Paper %T Minibatch Gibbs Sampling on Large Graphical Models %A Chris De Sa %A Vincent Chen %A Wing Wong %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-desa18a %I PMLR %P 1165--1173 %U https://proceedings.mlr.press/v80/desa18a.html %V 80 %X Gibbs sampling is the de facto Markov chain Monte Carlo method used for inference and learning on large scale graphical models. For complicated factor graphs with lots of factors, the performance of Gibbs sampling can be limited by the computational cost of executing a single update step of the Markov chain. This cost is proportional to the degree of the graph, the number of factors adjacent to each variable. In this paper, we show how this cost can be reduced by using minibatching: subsampling the factors to form an estimate of their sum. We introduce several minibatched variants of Gibbs, show that they can be made unbiased, prove bounds on their convergence rates, and show that under some conditions they can result in asymptotic single-update-run-time speedups over plain Gibbs sampling.
APA
De Sa, C., Chen, V. & Wong, W.. (2018). Minibatch Gibbs Sampling on Large Graphical Models. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:1165-1173 Available from https://proceedings.mlr.press/v80/desa18a.html.

Related Material