Marc Niethammer University of North Carolina Computer Science Image registration is an important component for many medical image analysis methods. It is a means to establish spatial correspondences within or across subjects. Non-parametric registration approaches typically involve the solution of very high-dimensional numerical optimization problems. However, more recently, deep learning approaches have been proposed that replace numerical optimization by regression. These approaches are therefore very fast at test time. In this talk I will give an overview of some work on deep-learning approaches for medical image registration, and, in particular, will talk about some recent work on metric estimation for image registration.