BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:2371b226fb70c80b63d7b00e04f7ac25
DTSTAMP:20260315T212046Z
SUMMARY:IDE Seminar Series – Dr Elizaveta Semenova
DESCRIPTION:How can deep generative modelling help with spatial statistics?
 \nGaussian processes (GPs)\, implemented through multivariate Gaussian dis
 tributions for a finite collection of data\, are the most popular approach
  in small-area spatial statistical modelling. In this context\, they are u
 sed to encode correlation structures over space and can generalize well in
  interpolation tasks. Despite their flexibility\, off-the-shelf GPs presen
 t serious computational challenges which limit their scalability and pract
 ical usefulness in applied settings. In this talk I will present a recentl
 y proposed novel approach\, using deep generative modelling to tackle this
  challenge\, termed PriorVAE: for a particular spatial setting\, we approx
 imate a class of GP priors through prior sampling and subsequent fitting o
 f a variational autoencoder (VAE). Given a trained VAE\, the resultant dec
 oder allows spatial inference to become incredibly efficient due to the lo
 w dimensional\, independently distributed latent Gaussian space representa
 tion of the VAE. Once trained\, inference using the VAE decoder replaces t
 he GP within a Bayesian sampling framework. This approach provides tractab
 le and easy-to-implement means of approximately encoding spatial priors an
 d facilitates efficient statistical inference. During the talk\, I will ex
 plain the methodology\, demonstrate its applications and discuss potential
  future impact of this research direction. 
URL:https://www.imperial.ac.uk/events/153996/ide-seminar-series-dr-elizavet
 a-semenova/
DTSTART;TZID=Europe/London:20221025T153000
DTEND;TZID=Europe/London:20221025T163000
LOCATION:Cockburn Lecture Theatre\, Medical School\, St Mary's Campus\, Imp
 erial College London\, London\, W2 1NY\, United Kingdom
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
DTSTART:20221025T153000
TZNAME:BST
TZOFFSETTO:+0100
TZOFFSETFROM:+0100
END:DAYLIGHT
END:VTIMEZONE
END:VCALENDAR