Topology-Preserving Prototype Learning on Riemannian Manifolds Konferenzpaper