Spectral clustering and approximation of the Laplace-Beltrami operator

Jeudi 28 novembre 2019 11:00-12:00 - Vincent Divol - Université Paris-Sud

Résumé : Clustering is one of the fundamental tasks of machine learning : it consists in classifying observations into different groups (or clusters) which are believed to have similar behaviors. Spectral clustering is one of the most widely-used clustering technique, making use of the Laplacian defined on an appropriate graph to separate clusters which might possess complex geometry. We will give some hindsight on this method, by considering it as a spectral estimation procedure of the Laplace-Beltrami operator on some unknown manifold underlying the observed data points.

Spectral clustering and approximation of the Laplace-Beltrami operator  Version PDF