Probabilistic programming for Java using symbolic execution and model counting

In this paper we describe a probabilistic programming environment for Java that is based on symbolic execution and model counting. The novelty of the framework is that the probability distributions in the program can themselves be symbolic, which allows parametric probabilistic programming. The framework handles typical probabilistic programming features, such as observe statements, and can be used for the encoding and analysis of Discrete Time Markov Chains (DTMC), Bayesian Networks, etc. We show two examples of using the system: (1) analysis of bubble sort when using an unreliable comparison operation, and, (2) analysis of a simulation model of autonomous aircraft towing vehicles, to show whether plans generated for these vehicles are robust when probability distributions are changed from the ones used to generate the plans.