Structures and Protocols for Inference

The widespread adoption of machine learning techniques is currently hindered by a lack of interoperability between the software toolkits implementing those techniques. We intend to address this problem by identifying a concise, expressive and extensible set of primitive structures and protocols for their combination that can be used to describe a wide variety of inference problems. These primitives will be implemented on existing open source and commercial toolkits allowing practitioners to solve inference problems by combining the strengths of these toolkits through this common interface.