Control-Data Separation and Logical Condition Propagation for Efficient Inference on Probabilistic Programs

TL;DR

This paper introduces a new sampling framework for probabilistic programs that combines control-data separation and logical condition propagation, significantly improving efficiency especially with complex loops and rare observations.

Contribution

It presents a novel combination of control-data separation and logical condition propagation for probabilistic inference, implemented on Anglican.

Findings

Enhanced efficiency for programs with while loops

Improved handling of rare observations

Effective integration of two recent ideas

Abstract

We present a novel sampling framework for probabilistic programs. The framework combines two recent ideas -- \emph{control-data separation} and \emph{logical condition propagation} -- in a nontrivial manner so that the two ideas boost the benefits of each other. We implemented our algorithm on top of Anglican. The experimental results demonstrate our algorithm's efficiency, especially for programs with while loops and rare observations.