Particle Gibbs with Ancestor Sampling for Probabilistic Programs

Jan-Willem Meent, Hongseok Yang, Vikash Mansinghka, Frank Wood
Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics, PMLR 38:986-994, 2015.

Abstract

Particle Markov chain Monte Carlo techniques rank among current state-of-the-art methods for probabilistic program inference. A drawback of these techniques is that they rely on importance resampling, which results in degenerate particle trajectories and a low effective sample size for variables sampled early in a program. We here develop a formalism to adapt ancestor resampling, a technique that mitigates particle degeneracy, to the probabilistic programming setting. We present empirical results that demonstrate nontrivial performance gains.

Cite this Paper

BibTeX
@InProceedings{pmlr-v38-vandemeent15, title = {{Particle Gibbs with Ancestor Sampling for Probabilistic Programs}}, author = {Meent, Jan-Willem and Yang, Hongseok and Mansinghka, Vikash and Wood, Frank}, booktitle = {Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics}, pages = {986--994}, year = {2015}, editor = {Lebanon, Guy and Vishwanathan, S. V. N.}, volume = {38}, series = {Proceedings of Machine Learning Research}, address = {San Diego, California, USA}, month = {09--12 May}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v38/vandemeent15.pdf}, url = {https://proceedings.mlr.press/v38/vandemeent15.html}, abstract = {Particle Markov chain Monte Carlo techniques rank among current state-of-the-art methods for probabilistic program inference. A drawback of these techniques is that they rely on importance resampling, which results in degenerate particle trajectories and a low effective sample size for variables sampled early in a program. We here develop a formalism to adapt ancestor resampling, a technique that mitigates particle degeneracy, to the probabilistic programming setting. We present empirical results that demonstrate nontrivial performance gains.} }
Endnote
%0 Conference Paper %T Particle Gibbs with Ancestor Sampling for Probabilistic Programs %A Jan-Willem Meent %A Hongseok Yang %A Vikash Mansinghka %A Frank Wood %B Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2015 %E Guy Lebanon %E S. V. N. Vishwanathan %F pmlr-v38-vandemeent15 %I PMLR %P 986--994 %U https://proceedings.mlr.press/v38/vandemeent15.html %V 38 %X Particle Markov chain Monte Carlo techniques rank among current state-of-the-art methods for probabilistic program inference. A drawback of these techniques is that they rely on importance resampling, which results in degenerate particle trajectories and a low effective sample size for variables sampled early in a program. We here develop a formalism to adapt ancestor resampling, a technique that mitigates particle degeneracy, to the probabilistic programming setting. We present empirical results that demonstrate nontrivial performance gains.
RIS
TY - CPAPER TI - Particle Gibbs with Ancestor Sampling for Probabilistic Programs AU - Jan-Willem Meent AU - Hongseok Yang AU - Vikash Mansinghka AU - Frank Wood BT - Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics DA - 2015/02/21 ED - Guy Lebanon ED - S. V. N. Vishwanathan ID - pmlr-v38-vandemeent15 PB - PMLR DP - Proceedings of Machine Learning Research VL - 38 SP - 986 EP - 994 L1 - http://proceedings.mlr.press/v38/vandemeent15.pdf UR - https://proceedings.mlr.press/v38/vandemeent15.html AB - Particle Markov chain Monte Carlo techniques rank among current state-of-the-art methods for probabilistic program inference. A drawback of these techniques is that they rely on importance resampling, which results in degenerate particle trajectories and a low effective sample size for variables sampled early in a program. We here develop a formalism to adapt ancestor resampling, a technique that mitigates particle degeneracy, to the probabilistic programming setting. We present empirical results that demonstrate nontrivial performance gains. ER -
APA
Meent, J., Yang, H., Mansinghka, V. & Wood, F.. (2015). Particle Gibbs with Ancestor Sampling for Probabilistic Programs. Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 38:986-994 Available from https://proceedings.mlr.press/v38/vandemeent15.html.

Related Material