Transforming Probabilistic Programs into Algebraic Circuits for Inference and Learning

Pedro Zuidberg Dos Martires, Vincent Derkinderen, Robin Manhaeve, Wannes Meert, Angelika Kimmig, Luc De Raedt

Keywords: arithmetic circuits, knowledge compilation, probabilistic programming, probabilistic inference

Abstract: Probabilistic (logic) programs are routinely compiled into arithmetic circuits. During such a compilation step, the logic representation of a probabilistic program is transformed into an arithmetic representation. We show that this transformation and the resulting circuits cannot only be used for discrete probabilistic inference, but also for a number of other tasks such as differentiation, learning and probabilistic inference in the discrete-continuous domain.