Séminaire de Probabilités et Statistique
lundi 12 mars 2007 à 10:30 - Cirad, amphi J. Alliot(Lavalette)
Susan Holmes (Univ Stanford USA - Univ Nice & INRA Mtp)
"Horseshoes and Kernel Methods (à la recherche du temps perdu)"
Multidimensional Scaling and Kernel methods can be used to find hidden gradients in multidimensional data, this the special case of manifold learning when the manifold is a curve. An often encountered pattern that ecologists know to look for is the "horseshoe" or "arch". I will show some examples of this phenomena in biology and political science. We can prove analytically in a simple case how this pattern occurs. This is joint work with Persi Diaconis and Sharad Goel. Pour plus d'informations, cliquez ici