BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:ML@CL Ad-hoc Seminar Series
SUMMARY:Constructing expressive models using nonparametric
  Gaussian process convolutions. - Magnus Ross\, Un
 iversity of Manchester
DTSTART;TZID=Europe/London:20230525T150000
DTEND;TZID=Europe/London:20230525T160000
UID:TALK201241AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/201241
DESCRIPTION:Gaussian processes (GPs) are powerful models for p
 robabilistic machine learning\, but good performan
 ce requires the selection of an appropriate covari
 ance function. Process convolutions are framework 
 for the construction of flexible covariances for b
 oth scalar and vector valued GPs\, but still requi
 re choosing a smoothing kernel\, which is non-triv
 ial. Previous approaches have built covariance fun
 ctions by using GP priors over the smoothing kerne
 l\, and by extension the covariance\, as a way to 
 bypass the need to specify it in advance. However\
 , such models have been limited in several ways: t
 hey are restricted to single dimensional inputs\, 
 e.g. time\; they only allow modeling of single out
 puts and they do not scale to large datasets since
  inference is not straightforward. In this talk I 
 will discuss the development of sampling and infer
 ence methods for nonparametric process convolution
 s that resolve these limitations\, and allow covar
 iances to be inferred from data in a large variety
  of domains. Using these methods\, I'll discuss so
 me extensions of process convolutions framework to
  the case of a nonlinear operators via the Volterr
 a series\, which result in non-Gaussian output pro
 cesses\, and allow for the development of powerful
  models for system identification. I will end by d
 iscussing the limitations of the framework\, some 
 ideas for future work. \n\nBio\n---\nMagnus Ross i
 s a third-year PhD student at the University of Ma
 nchester (and previously the University of Sheffie
 ld)\, working with Mauricio Álvarez and Michael Sm
 ith. His research interests center around Gaussian
  processes\, and in particular their use for physi
 cs informed machine learning.
LOCATION:Computer Lab\, SW00
CONTACT:Carl Henrik Ek
END:VEVENT
END:VCALENDAR