Probabilistic Numerics (Philipp Hennig)
Emmy Noether Research Group
The research group studies and develops numerical methods for use in intelligent systems. At the core of our approach lies the observation that algorithms for the computation of intractable objects, like integrals and extrema, can be phrased as active inference, as learning machines. It is thus possible to adapt existing numerical methods to specifically model and leverage the helpful structure and complicating sources of uncertainty that arise in intelligent systems, to deliver more robust or efficient answers.