---

Mercredi 14 mai 2008, 14h30, Salle S16 (Passage Saumon, niveau -1)

• Orateur/Speaker : John Harer, Duke University
• Titre/Title : Persistent homology - Finding global shape in datasets
• Résumé/Abstract :

  ---

  Dimension reduction for scientific datasets is an important and difficult problem. One aspect, called manifold learning, is a subject of growing interest in the statistical and computer science communities where a variety of methods exist for recognizing approximate local dependency in data. In this talk we will discuss a new method based on Persistent Local Homology. We will be especially interested in the question of when a data set has the structure that one would expect from a stratified space, which is a collection of manifolds assembled in a particular way.

  ---

  In this talk we will define persistence, persistent local homology and stratified spaces and show how they can be used to find interesting structures. This work is joint with Paul Bendich, Herbert Edelsbrunner and Dmitriy Morozov.