Functoriality in Geometric Data

This report provides an overview of the talks at the Dagstuhl Seminar 17021 “Functoriality in Geometric Data”. The seminar brought together researchers interested in the fundamental questions of similarity and correspondence across geometric data sets, which include collections of GPS traces, images, 3D shapes and other types of geometric data. A recent trend, emerging independently in multiple theoretical and applied communities, is to understand networks of geometric data sets through their relations and interconnections, a point of view that can be broadly described as exploiting the functoriality of data, which has a long tradition associated with it in mathematics. Functoriality, in its broadest form, is the notion that in dealing with any kind of mathematical object, it is at least as important to understand the transformations or symmetries possessed by the object or the family of objects to which it belongs, as it is to study the object itself. This general idea has led to deep insights into the structure of various geometric spaces as well as to state-of-the-art methods in various application domains. The talks spanned a wide array of subjects under the common theme of functoriality, including: the analysis of geometric collections, optimal transport for geometric datasets, deep learning applications and many more. Seminar January 8–13, 2017 – http://www.dagstuhl.de/17021 1998 ACM Subject Classification I.3.5 Computational Geometry and Object Modeling, I.4 Image Processing and Computer Vision