BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Fast Gaussian process learning for regression\, semi-supervised cl
 assification\, and multiway analysis - Prof Alan Qi (Purdue U)
DTSTART:20120704T130000Z
DTEND:20120704T140000Z
UID:TALK38815@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:In this talk I will cover two topics on Gaussian process learn
 ing. First\,\n      I will describe tensor-variate latent nonparametric Ba
 yesian\n      models\, coupled with efficient inference methods\, for mult
 iway\n      data analysis. Unlike classical tensor decomposition models\, 
 our\n      new approaches model nonlinear interactions and handle both\n  
     continuous and binary data. To efficiently learn the InfTucker\n      
 from data\, we develop a variational inference technique on\n      tensors
 . Compared with classical implementation\, the new technique  reduces both
  time and space complexities by several orders of  magnitude. Our experime
 ntal results on chemometrics and social\n      network datasets demonstrat
 e that our new models can achieve\n      significantly higher prediction a
 ccuracy than state-of-art tensor\n      decomposition approaches. Furtherm
 ore\, for two dimensional\n      problems\, our tensor model reduces to no
 nlinear stochastic\n      blockmodels for network modeling\, which I will 
 briefly discuss in\n      the talk as well. Second\, I will describe a new
  sparse Gaussian\n      process model\, EigenGP\, based on Karhunen-Loeve 
 (KL) expansions of\n      a GP prior. We can view this new approach as spa
 rse PCA in a\n      functional space\, which not only reduces the computat
 ional cost of\n      GP inference but also has the potential of further im
 proving the\n      predictive performance of a full GP. By selecting eigen
 functions\n      of Gaussian kernels that are associated with data cluster
 s\,\n      EigenGP is also suitable for semi-supervised learning. Our\n   
    experimental results demonstrate improved predictive performance\n     
  of EigenGP over several state-of-the- art sparse GP and\n      semisuperv
 ised learning methods for regression\, classification\,\n      and semisup
 ervised classification. 
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR