Title: Differentially Private Bayesian Programming

Abstract: We present an expressive framework, called PrivInfer, for writing and verifying differentially private machine learning algorithms. Programs in PrivInfer are written in a rich functional probabilistic language with constructs for performing Bayesian inference. Then, differential privacy of programs is established using a relational refinement type system, in which refinements on probability types are indexed by a metric on distributions. Our framework leverages recent developments in Bayesian inference, probabilistic program- ming languages, and in relational refinement types. We demonstrate the expressiveness of PrivInfer by verifying privacy for several examples of private Bayesian inference.