I have left the MPI. Please find my new homepage here.

• Identification of differential equation systems from data samples. I have investigated the (partial) equivalence between kernel methods such as SVMs or Gaussian Processes and well-known system identification tools such as the Kalman filter and prediction error methods.
• Estimation of gene regulatory networks from micro-array data. We use Bayesian inference and experimental design in the sparse linear, a special kind of dynamical system identification.
• Structured/manifold-valued output. It is noteworthy that regression from structured domains to linear spaces is quite well understood, and kernel machines have been adapted to this setting. On the contrary, regression from a linear space to a structured domain, is much less well explored.
• Machine learning in computer graphics. The work focused on implicit surface reconstruction using SVMs and on reformulating dense 3D correspondence in terms of a machine learning problem.