Slicing of Probabilistic Programs: A Review of Existing Approaches

Abstract

Program slicing aims to simplify programs by identifying and removing non-essential parts while preserving program behavior. It is widely used for program understanding, debugging, and software maintenance. This article provides an overview of slicing techniques for probabilistic programs, which blend traditional programming constructs with random sampling and conditioning. These programs have experienced a notable resurgence in recent years due to new applications in fields such as artificial intelligence and differential privacy. Concretely, we review the three major slicing techniques currently available for probabilistic programs: the foundational technique by Hur et al., the subsequent development by Amtoft and Banerjee based on probabilistic control flow graphs, and the more recent approach by Navarro and Olmedo based on program specifications. We provide a clear, accessible, and self-contained presentation of these techniques, and compare them across multiple dimensions to provide a deeper insight into the current state-of-the-art in probabilistic program slicing.