Agenda

Andrea Gasparetto, DAIS

Abstract:
While there exists a huge framework of techniques that deal with classification and regression with real-valued data which can be embedded into the Euclidean vector space, dealing with data that lay in different spaces is another matter. We show how data that are defined in a non-vectorial space can be processed in order to tackle problem like data clustering and classification. We present two methods that are based on spectral descriptors. In the first one we lift the one-to-one node-correspondence assumption. In the second approach, in which we study the variations of a metric defined on a special manifold, we deal with the problem of missing correspondences casting it as a bipartite matching problem. Finally, we present a technique based on functional correspondences that computes an injective map between two surfaces.