BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Graphs\, Cuts and p-Spectral Clustering
DTSTART:20090504T161500
DTSTAMP:20260408T141719Z
UID:7426439061bec7a6c9f4f418efa7f0495e4e54e43475b59fec84fede
CATEGORIES:Conferences - Seminars
DESCRIPTION:Dr. Matthias Hein\, Saarland University\, Germany\nGraph-based
  methods can be applied to any kind of data and are\ntherefore heavily use
 d in practice. In machine learning most of\nthe time so called similarity 
 graphs are used. Although the\npractical results of learning algorithms de
 pend heavily on the\ngraph construction\, this is a largely unexplored are
 a in machine\nlearning. In this talk I show results from manifold learning
  and\nclustering which illustrate the influence of graph type and graph\np
 arameters. In particular\, I present recent results which show\nthat the p
 opulation version of the clustering objective induced\nby the normalized c
 ut criterion depends on the employed graph type.\n\nIn the second half of 
 the talk I discuss a generalized version of\nspectral clustering based on 
 eigenvectors of the graph\np-Laplacian\, a non-linear generalization of th
 e graph Laplacian.\nInterestingly\, one can prove that the cut value obtai
 ned by\nthresholding the second eigenvector of the p-Laplacian converges\n
 towards the optimal Cheeger cut as p tends to 1.\n\nBio: Matthias Hein has
  been researcher from 2002 to 2007 at the\nMax-Planck-Institute for Biolog
 ical Cybernetics in the Empirical\nInference group of Prof. Schoelkopf. He
  received his doctoral\ndegree in Computer Science in 2005 from the Techni
 cal University\nDarmstadt. Since 2007 he is Juniorprofessor at the Compute
 r\nScience Department at Saarland University.\n\nM. Hein's homepage
LOCATION:INM202
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR