Kernel methods on Riemannian manifolds

Transcription

Kernel methods on Riemannian manifolds
Kernel methods on Riemannian manifolds Richard Hartley, Sadeep Jayasumana, Mehrtash Harandi, Mathieu Salzmann Hongdong Li, Khurrum A>ab Toulouse, Jan 2014 James Chester Manifold: prominent Australian poliGcian and grazier. graze (v.1)
"to feed," Old English grasian "to feed on grass," from græs "grass"
(see grass). Cf. Middle Dutch, Middle High German grasen, Dutch grazen,
German grasen. French gazon. Figurative use by 1570s. Related: Grazed;
grazing.
Perhaps the first Grass-­‐man Perhaps the first Manifold Grass-­‐man Manifold Image from hNp://en.wikipedia.org/wiki/
File:TangenGalvektor.svg ExponenGal map. Set out in a given direcGon and see where you get to. Logarithm map. What direcGon do I need to go to get from X to Y ? x y Kernels on PosiGve Definite Matrices Example: Kernel DicGonary Learning on Manifolds Harandi et al (ICCV 2013) Planes through the origin ProjecGve space – lines through the origin Atul Kanaujia Shape Manifold. Captures the configuraGon of a set of points, allowing for rotaGon, translaGon and scaling. Guillaume Charpiat The Hopf FibraGon – the realizaGon of CP1 as a fibraGon Leaf Database Acknowledgements. This talk deals with work done by myself and my collaborators, parGcularly 1.  Mehrtash Harandi 2.  Sadeep Jayasumana 3.  Mathieu Salzmann 4.  Hongdong Li 5.  Brian Lovell The End