Encapsulating models and approximate inference programs in probabilistic modules
This work addresses the challenge of integrating diverse probabilistic models and inference methods in a modular way for researchers and practitioners in machine learning and statistics.
The paper tackles the problem of encapsulating complex probabilistic models with latent variables and custom approximate inference machinery by introducing a probabilistic module interface, which provides a platform-agnostic abstraction barrier and enables sound approximate inference algorithms for networks of such modules.
This paper introduces the probabilistic module interface, which allows encapsulation of complex probabilistic models with latent variables alongside custom stochastic approximate inference machinery, and provides a platform-agnostic abstraction barrier separating the model internals from the host probabilistic inference system. The interface can be seen as a stochastic generalization of a standard simulation and density interface for probabilistic primitives. We show that sound approximate inference algorithms can be constructed for networks of probabilistic modules, and we demonstrate that the interface can be implemented using learned stochastic inference networks and MCMC and SMC approximate inference programs.