Stable and interpretable topological feature maps
Martina Scolamiero (KTH Royal Institute of Technology - Sweden)
Abstract: Persistent homology, a popular method in TDA, can be used to define feature maps encoding geometrical properties of data. In this talk I will present a method, developed by the TDA group at KTH, which allows to construct feature maps with learnable parameters, stable with respect to distances on persistence modules. The feature maps are in fact defined starting from distances between persistence modules rather than on the barcode decomposition, making the method suitable for generalisations. Particular focus will be on understanding parametrised families of such feature maps, such as those stable with respect to p-Wasserstein distance. The use of Wasserstein stable features will be illustrated on real world and artificial datasets.
geometric topology
Audience: researchers in the topic
Series comments: Web-seminar series on Applications of Geometry and Topology
Organizers: | Alicia Dickenstein, José-Carlos Gómez-Larrañaga, Kathryn Hess, Neza Mramor-Kosta, Renzo Ricca*, De Witt L. Sumners |
*contact for this listing |