The goal is to automate probabilistic modeling of dynamical systems (and their surroundings) via a formally defined probabilistic modeling language. We have identified specific intermediate research goals and organized these into three disciplinary research themes.

Research Theme 1: Develop a formally defined probabilistic modeling language and an efficient model compiler that is directly suitable for industrial adoption. For a pre-release of our probabilistic programming language Birch, click here.

Research Theme 2: Construct probabilistic models representing complex dynamical systems that gain situational awareness in their environments using high-dimensional sensor data to automatically compute system controllers, obviating the need for explicit tuning by a machine learning expert.

Research Theme 3: Automate complexity reducing techniques for learning in high-dimensional models, with the goal of making the methods viable for inclusion in the model compiler.