Hierarchical Reasoning with Probabilistic Programming

Brian E. Ruttenberg, Matthew P. Wilkins, Avi Pfeffer

Abstract: Hierarchical representations are common in many artificial intelligence tasks, such as classification of satellites in orbit. Representing and reasoning on hierarchies is difficult, however, as they can be large, deep and constantly evolving. Although probabilistic programming provides the flexibility to model many situations, current probabilistic programming languages (PPL) do not adequately support hierarchical reasoning. We present a novel PPL approach to representing and reasoning about hierarchies that utilizes references, enabling unambiguous access and referral to hierarchical objects and their properties.