Ensuring Rapid Mixing and Low Bias for Asynchronous Gibbs Sampling

Christopher De Sa, Chris Re, Kunle Olukotun
Proceedings of The 33rd International Conference on Machine Learning, PMLR 48:1567-1576, 2016.

Abstract

Gibbs sampling is a Markov chain Monte Carlo technique commonly used for estimating marginal distributions. To speed up Gibbs sampling, there has recently been interest in parallelizing it by executing asynchronously. While empirical results suggest that many models can be efficiently sampled asynchronously, traditional Markov chain analysis does not apply to the asynchronous case, and thus asynchronous Gibbs sampling is poorly understood. In this paper, we derive a better understanding of the two main challenges of asynchronous Gibbs: bias and mixing time. We show experimentally that our theoretical results match practical outcomes.

Cite this Paper


BibTeX
@InProceedings{pmlr-v48-sa16, title = {Ensuring Rapid Mixing and Low Bias for Asynchronous Gibbs Sampling}, author = {Sa, Christopher De and Re, Chris and Olukotun, Kunle}, booktitle = {Proceedings of The 33rd International Conference on Machine Learning}, pages = {1567--1576}, year = {2016}, editor = {Balcan, Maria Florina and Weinberger, Kilian Q.}, volume = {48}, series = {Proceedings of Machine Learning Research}, address = {New York, New York, USA}, month = {20--22 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v48/sa16.pdf}, url = { http://proceedings.mlr.press/v48/sa16.html }, abstract = {Gibbs sampling is a Markov chain Monte Carlo technique commonly used for estimating marginal distributions. To speed up Gibbs sampling, there has recently been interest in parallelizing it by executing asynchronously. While empirical results suggest that many models can be efficiently sampled asynchronously, traditional Markov chain analysis does not apply to the asynchronous case, and thus asynchronous Gibbs sampling is poorly understood. In this paper, we derive a better understanding of the two main challenges of asynchronous Gibbs: bias and mixing time. We show experimentally that our theoretical results match practical outcomes.} }
Endnote
%0 Conference Paper %T Ensuring Rapid Mixing and Low Bias for Asynchronous Gibbs Sampling %A Christopher De Sa %A Chris Re %A Kunle Olukotun %B Proceedings of The 33rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2016 %E Maria Florina Balcan %E Kilian Q. Weinberger %F pmlr-v48-sa16 %I PMLR %P 1567--1576 %U http://proceedings.mlr.press/v48/sa16.html %V 48 %X Gibbs sampling is a Markov chain Monte Carlo technique commonly used for estimating marginal distributions. To speed up Gibbs sampling, there has recently been interest in parallelizing it by executing asynchronously. While empirical results suggest that many models can be efficiently sampled asynchronously, traditional Markov chain analysis does not apply to the asynchronous case, and thus asynchronous Gibbs sampling is poorly understood. In this paper, we derive a better understanding of the two main challenges of asynchronous Gibbs: bias and mixing time. We show experimentally that our theoretical results match practical outcomes.
RIS
TY - CPAPER TI - Ensuring Rapid Mixing and Low Bias for Asynchronous Gibbs Sampling AU - Christopher De Sa AU - Chris Re AU - Kunle Olukotun BT - Proceedings of The 33rd International Conference on Machine Learning DA - 2016/06/11 ED - Maria Florina Balcan ED - Kilian Q. Weinberger ID - pmlr-v48-sa16 PB - PMLR DP - Proceedings of Machine Learning Research VL - 48 SP - 1567 EP - 1576 L1 - http://proceedings.mlr.press/v48/sa16.pdf UR - http://proceedings.mlr.press/v48/sa16.html AB - Gibbs sampling is a Markov chain Monte Carlo technique commonly used for estimating marginal distributions. To speed up Gibbs sampling, there has recently been interest in parallelizing it by executing asynchronously. While empirical results suggest that many models can be efficiently sampled asynchronously, traditional Markov chain analysis does not apply to the asynchronous case, and thus asynchronous Gibbs sampling is poorly understood. In this paper, we derive a better understanding of the two main challenges of asynchronous Gibbs: bias and mixing time. We show experimentally that our theoretical results match practical outcomes. ER -
APA
Sa, C.D., Re, C. & Olukotun, K.. (2016). Ensuring Rapid Mixing and Low Bias for Asynchronous Gibbs Sampling. Proceedings of The 33rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 48:1567-1576 Available from http://proceedings.mlr.press/v48/sa16.html .

Related Material