BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Extreme Partial Least Squares
DTSTART:20260112T123000Z
DTEND:20260112T133000Z
DTSTAMP:20260424T050800Z
UID:indico-event-14707@indico.math.cnrs.fr
DESCRIPTION:Speakers: Stéphane GIRARD\n\nThe talk deals with dimension-re
 duction techniques for modelling conditional extreme values. Specifically\
 , we investigate the idea that extreme values of a response variable can b
 e explained by nonlinear functions derived from linear projections of an i
 nput random vector. In this context\, the estimation of projection directi
 ons is examined and the Extreme Partial Least Squares (EPLS) method is int
 roduced as an adaptation of the original Partial Least Squares (PLS) metho
 d tailored to the extreme-value framework. Further\, an interpretation of 
 EPLS directions as maximum likelihood estimators is proposed\, utilizing t
 he von Mises--Fisher distribution applied to hyperballs. The dimension red
 uction process is enhanced through the Bayesian paradigm\, enabling the in
 corporation of prior information into the projection direction estimation.
  The maximum a posteriori estimator is derived in two specific cases\, elu
 cidating it as a regularization or shrinkage of the EPLS estimator. We als
 o establish its asymptotic behavior as the sample size approaches infinity
 . A simulation data study is conducted in order to assess the practical ut
 ility of our proposed method. This clearly demonstrates its effectiveness 
 even in moderate data problems within high-dimensional settings. Furthermo
 re\, we provide an illustrative example of the method's applicability usin
 g French farm income data\, highlighting its efficacy in real-world scenar
 ios.\nThis is joint work with J. Arbel\, M. Bousebata (Inria)\, H. Lorenzo
  (Univ. Aix-Marseille) and C. Pakzad (Univ. Paris-Nanterre).\n\nhttps://in
 dico.math.cnrs.fr/event/14707/
LOCATION:Fokko du Cloux (La Doua)
URL:https://indico.math.cnrs.fr/event/14707/
END:VEVENT
END:VCALENDAR