BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Shared Stochastic Gaussian Process Latent Variable Models: A Multi
 -modal Generative Model for Quasar Spectra - Dr. Vidhi Lalchand (MIT)
DTSTART:20250324T160000Z
DTEND:20250324T170000Z
UID:TALK229201@talks.cam.ac.uk
CONTACT:65128
DESCRIPTION:This work proposes a scalable probabilistic latent variable mo
 del based on Gaussian processes\n(Lawrence\, 2004) in the context of multi
 ple observation spaces. We focus on an application\nin astrophysics where 
 it is typical for data sets to contain both observed spectral features\nas
  well as scientific properties of astrophysical objects such as galaxies o
 r exoplanets. In\nour application\, we study the spectra of very luminous 
 galaxies known as quasars\, and their\nproperties\, such as the mass of th
 eir central supermassive black hole\, their accretion rate\nand their lumi
 nosity\, and hence\, there can be multiple observation spaces. A single da
 ta\npoint is then characterised by different classes of observations\, whi
 ch may have different\nlikelihoods. Our proposed model extends the baselin
 e stochastic variational Gaussian\nprocess latent variable model (GPLVM) (
 Lalchand et al.\, 2022) to this setting\, proposing a\nseamless generative
  model where the quasar spectra and scientific labels can be generated\nsi
 multaneously when modelled with a shared latent space acting as input to d
 ifferent sets\nof Gaussian process decoders\, one for each observation spa
 ce. In addition\, this framework\nallows training in the missing data sett
 ing where a large number of dimensions per data\npoint may be unknown or u
 nobserved. We demonstrate high-fidelity reconstructions of the\nspectra an
 d the scientific labels during test-time inference and briefly discuss the
  scientific\ninterpretations of the results along with the significance of
  such a generative model.
LOCATION:Martin Ryle Seminar Room\, KICC
END:VEVENT
END:VCALENDAR