Ohio State nav bar

Topology, Geometry and Data Seminar - Mathieu Carrière

Topology, Geometry and Data Seminar
April 23, 2019
4:10PM - 5:10PM
Cockins Hall 240

Date Range
Add to Calendar 2019-04-23 16:10:00 2019-04-23 17:10:00 Topology, Geometry and Data Seminar - Mathieu Carrière Title: Kernel  Methods and Persistence Diagrams Speaker: Mathieu Carrière (Columbia University) Abstract: Persistence Diagrams (PDs) are important feature descriptors in Topological Data Analysis. Due to the nonlinearity of the space of PDs equipped with their diagram distances, most of the recent attempts at using PDs in Machine Learning have been done through kernel methods, i.e., embeddings of PDs into Reproducing Kernel Hilbert Spaces (RKHS), in which all computations can be performed easily. Since PDs enjoy theoretical stability guarantees for the diagram distances, the metric properties of a kernel k, i.e., the relationship between the RKHS distance dk and the diagram distances, are of central interest for understanding if the PD guarantees carry over to the embedding. In this talk, I will present different ways to embed PDs into RKHSs, and study the possibility of embedding PDs into general RKHS with bi-Lipschitz maps. In particular, we show that when the RKHS is infinite dimensional, any lower bound must depend on the cardinalities of the PDs, and that when the RKHS is finite dimensional, finding a bi-Lipschitz embedding is impossible, even when restricting the PDs to have bounded cardinalities. Seminar URL: https://tgda.osu.edu/ Cockins Hall 240 Department of Mathematics math@osu.edu America/New_York public

Title: Kernel  Methods and Persistence Diagrams

SpeakerMathieu Carrière (Columbia University)

Abstract: Persistence Diagrams (PDs) are important feature descriptors in Topological Data Analysis. Due to the nonlinearity of the space of PDs equipped with their diagram distances, most of the recent attempts at using PDs in Machine Learning have been done through kernel methods, i.e., embeddings of PDs into Reproducing Kernel Hilbert Spaces (RKHS), in which all computations can be performed easily. Since PDs enjoy theoretical stability guarantees for the diagram distances, the metric properties of a kernel k, i.e., the relationship between the RKHS distance dk and the diagram distances, are of central interest for understanding if the PD guarantees carry over to the embedding.

In this talk, I will present different ways to embed PDs into RKHSs, and study the possibility of embedding PDs into general RKHS with bi-Lipschitz maps. In particular, we show that when the RKHS is infinite dimensional, any lower bound must depend on the cardinalities of the PDs, and that when the RKHS is finite dimensional, finding a bi-Lipschitz embedding is impossible, even when restricting the PDs to have bounded cardinalities.

Seminar URLhttps://tgda.osu.edu/

Events Filters: