Learning proposals for probabilistic programs with inference combinators

Sam Stites, Heiko Zimmermann, Hao Wu, Eli Sennesh, Jan-Willem van de Meent
Proceedings of the Thirty-Seventh Conference on Uncertainty in Artificial Intelligence, PMLR 161:1056-1066, 2021.

Abstract

We develop operators for construction of proposals in probabilistic programs, which we refer to as inference combinators. Inference combinators define a grammar over importance samplers that compose primitive operations such as application of a transition kernel and importance resampling. Proposals in these samplers can be parameterized using neural networks, which in turn can be trained by optimizing variational objectives. The result is a framework for user-programmable variational methods that are correct by construction and can be tailored to specific models. We demonstrate the flexibility of this framework by implementing advanced variational methods based on amortized Gibbs sampling and annealing.

Cite this Paper

BibTeX
@InProceedings{pmlr-v161-stites21a, title = {Learning proposals for probabilistic programs with inference combinators}, author = {Stites, Sam and Zimmermann, Heiko and Wu, Hao and Sennesh, Eli and van de Meent, Jan-Willem}, booktitle = {Proceedings of the Thirty-Seventh Conference on Uncertainty in Artificial Intelligence}, pages = {1056--1066}, year = {2021}, editor = {de Campos, Cassio and Maathuis, Marloes H.}, volume = {161}, series = {Proceedings of Machine Learning Research}, month = {27--30 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v161/stites21a/stites21a.pdf}, url = {https://proceedings.mlr.press/v161/stites21a.html}, abstract = {We develop operators for construction of proposals in probabilistic programs, which we refer to as inference combinators. Inference combinators define a grammar over importance samplers that compose primitive operations such as application of a transition kernel and importance resampling. Proposals in these samplers can be parameterized using neural networks, which in turn can be trained by optimizing variational objectives. The result is a framework for user-programmable variational methods that are correct by construction and can be tailored to specific models. We demonstrate the flexibility of this framework by implementing advanced variational methods based on amortized Gibbs sampling and annealing.} }
Endnote
%0 Conference Paper %T Learning proposals for probabilistic programs with inference combinators %A Sam Stites %A Heiko Zimmermann %A Hao Wu %A Eli Sennesh %A Jan-Willem van de Meent %B Proceedings of the Thirty-Seventh Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2021 %E Cassio de Campos %E Marloes H. Maathuis %F pmlr-v161-stites21a %I PMLR %P 1056--1066 %U https://proceedings.mlr.press/v161/stites21a.html %V 161 %X We develop operators for construction of proposals in probabilistic programs, which we refer to as inference combinators. Inference combinators define a grammar over importance samplers that compose primitive operations such as application of a transition kernel and importance resampling. Proposals in these samplers can be parameterized using neural networks, which in turn can be trained by optimizing variational objectives. The result is a framework for user-programmable variational methods that are correct by construction and can be tailored to specific models. We demonstrate the flexibility of this framework by implementing advanced variational methods based on amortized Gibbs sampling and annealing.
APA
Stites, S., Zimmermann, H., Wu, H., Sennesh, E. & van de Meent, J.. (2021). Learning proposals for probabilistic programs with inference combinators. Proceedings of the Thirty-Seventh Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 161:1056-1066 Available from https://proceedings.mlr.press/v161/stites21a.html.

Related Material