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:Cosmology Lunch
SUMMARY:Nornalizing Flows for cosmology applications - Uro
s Seljak (Berkeley)
DTSTART;TZID=Europe/London:20211025T160000
DTEND;TZID=Europe/London:20211025T170000
UID:TALK162241AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/162241
DESCRIPTION:Normalizing Flows (NF) are bijective maps from the
data to a Gaussian (normal) distribution or vicev
ersa. In contrast to other generative models \nthe
y are lossless and provide data likelihood via the
Jacobian of the transformation. I will first pres
ent a novel Sliced Iterative NF (SINF)\, \nwhich i
s based on Optimal Transport theory\, achieving st
ate of the art results in density estimation for s
mall data samples and in anomaly detection applica
tions in high energy physics. \nI will discuss its
applications to Bayesian Inference and to Global
Optimization problems\, where it enables new metho
ds of sampling and optimization\, which have the p
otential to accelerate standard Monte Carlo Markov
Chains. In the second half of the talk I will pre
sent a Normalizing Flow for data structures with R
otational and Translational Equivariance (TRENF)\
, which can be used for generative modeling and li
kelihood analysis of cosmological data. By trainin
g the data likelihood on the posterior this approa
ch enables near optimal cosmological likelihood an
alysis\, where information from all the data is op
timally combined into a single number (likelihood)
as a function of cosmological parameters. This me
thod provides uncertainty quantification via the f
ull posterior of cosmological parameters\, which p
aves the way for a complete and optimal cosmologic
al data analysis with Normalizing Flows.
LOCATION:CMS\, Pav. B\, CTC Common Room (B1.19) [Potter Roo
m]
CONTACT:James Bonifacio
END:VEVENT
END:VCALENDAR