Composable inference metaprogramming using subproblems

Abstract

Inference metaprogramming enables effective probabilistic programming by supporting the decomposition of executions of probabilistic programs into subproblems and the deployment of hybrid probabilistic inference algorithms that apply different base probabilistic inference algorithms to different subproblems. I present the first sound and complete technique for extracting and stitching otherwise entangled subproblems for independent inference. I also prove asymptotic convergence results for hybrid inference algorithms for subproblem inference in probabilistic programs.