arXiv:2508.20922 · v3
Static Factorisation of Probabilistic Programs With User-Labelled Sample Statements and While Loops
摘要
It is commonly known that any Bayesian network can be implemented as a probabilistic program, but the reverse direction is not so clear. In this work, we address the open question to what extent a probabilistic program with user-labelled sample statements and while loops - features found in languages like Gen, Turing, and Pyro - can be represented graphically. To this end, we extend existing operational semantics to support these language features. By translating a program to its control-flow graph, we define a sound static analysis that approximates the dependency structure of the random variables in the program. As a result, we obtain a static factorisation of the implicitly defined program density, which is equivalent to the known Bayesian network factorisation for programs without loops and constant labels, but constitutes a novel graphical representation for programs that define an unbounded number of random variables via loops or dynamic labels. We further develop a sound program slicing technique to leverage this structure to statically enable three well-known optimisations for the considered program class: we reduce the variance of gradient estimates in variational inference and we speed up both single-site Metropolis Hastings and sequential Monte Carlo. These optimisations are proven correct and empirically shown to match or outperform existing techniques.
作者备注：v3: adds errata correcting Corollary 3; main results unaffected
期刊信息：Proc. ACM Program. Lang., Vol. 10, No. OOPSLA1, Article 115 (April 2026)