Towards representation agnostic probabilistic programming

TL;DR

This paper proposes a universal factor abstraction that enables representation-agnostic probabilistic programming, allowing flexible mixing of different model representations and facilitating inference in complex hybrid models.

Contribution

Introduction of a factor abstraction with five operations that serves as a universal interface for various factor representations in probabilistic programming.

Findings

Enables mixing of discrete, Gaussian, and sample-based representations.

Facilitates inference in complex hybrid models.

Provides a unified framework for diverse factor manipulations.

Abstract

Current probabilistic programming languages and tools tightly couple model representations with specific inference algorithms, preventing experimentation with novel representations or mixed discrete-continuous models. We introduce a factor abstraction with five fundamental operations that serve as a universal interface for manipulating factors regardless of their underlying representation. This enables representation-agnostic probabilistic programming where users can freely mix different representations (e.g. discrete tables, Gaussians distributions, sample-based approaches) within a single unified framework, allowing practical inference in complex hybrid models that current toolkits cannot adequately express.