Towards representation agnostic probabilistic programming
This addresses a bottleneck for researchers and practitioners in probabilistic programming by enabling more flexible model experimentation and inference in hybrid settings.
The paper tackles the problem of probabilistic programming languages coupling model representations with specific inference algorithms, which restricts experimentation with novel representations and mixed discrete-continuous models. It introduces a factor abstraction with five fundamental operations as a universal interface, enabling representation-agnostic probabilistic programming that allows mixing different representations in a unified framework for practical inference in complex hybrid models.
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.