Scalable Inference and Learning for Probabilistic Programs

ProbLog is an extension of Prolog that allows facts to be annotated with probabilities, defining a distribution over the possible worlds. Probabilistic inference is performed by grounding the program, encoding it in a logic formula, optionally compiling it into a more compact form, and finally performing a weighted model count on it to obtain an answer. Each of these steps still has room for improvement in terms of speed and efficiency. The aim of this research is to find better algorithms or implementations that perform these operations.