Amortized Rejection Sampling in Universal Probabilistic Programming
Machine Learning
2022-03-30 v3 Artificial Intelligence
Machine Learning
Abstract
Naive approaches to amortized inference in probabilistic programs with unbounded loops can produce estimators with infinite variance. This is particularly true of importance sampling inference in programs that explicitly include rejection sampling as part of the user-programmed generative procedure. In this paper we develop a new and efficient amortized importance sampling estimator. We prove finite variance of our estimator and empirically demonstrate our method's correctness and efficiency compared to existing alternatives on generative programs containing rejection sampling loops and discuss how to implement our method in a generic probabilistic programming framework.
Cite
@article{arxiv.1910.09056,
title = {Amortized Rejection Sampling in Universal Probabilistic Programming},
author = {Saeid Naderiparizi and Adam Ścibior and Andreas Munk and Mehrdad Ghadiri and Atılım Güneş Baydin and Bradley Gram-Hansen and Christian Schroeder de Witt and Robert Zinkov and Philip H. S. Torr and Tom Rainforth and Yee Whye Teh and Frank Wood},
journal= {arXiv preprint arXiv:1910.09056},
year = {2022}
}
Comments
AISTATS 2022 camera ready