BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:4
UID:UW-Physics-Event-6527
DTSTART:20211006T160000Z
DTEND:20211006T171500Z
DTSTAMP:20260414T192152Z
LAST-MODIFIED:20211005T035420Z
LOCATION:Chamberlin 5280 (Zoom link also available for online particip
 ants who signed up on our mailing list)
SUMMARY:Normalizing Flows for scientific applications\, Physics ∩ ML
  Seminar\, Uros Seljak\, UC Berkeley
DESCRIPTION:Normalizing Flows (NF) are bijective maps from the data to
  a Gaussian (normal) distribution or viceversa. In contrast to other g
 enerative models 
\nthey are lossless and provide data likelihood via 
 the Jacobian of the transformation. I will first present a novel Slice
 d Iterative NF (SINF)\, 
\nwhich is based on Optimal Transport theory\
 , achieving state of the art results in density estimation for small d
 ata samples and in anomaly detection applications in high energy physi
 cs. 
\nI will discuss its applications to Bayesian Inference and to Gl
 obal Optimization problems\, where it enables new methods of sampling 
 and optimization\, which have the potential to accelerate standard MCM
 C. In the second half of the talk I will present a Normalizing Flow fo
 r data structures with Rotational and Translational Equivariance  (TRE
 NF)\, which can be used for generative modeling and likelihood analysi
 s of cosmological data. By training the data likelihood on the posteri
 or this approach enables near optimal cosmological likelihood analysis
 \, where information from all the data is optimally combined into a si
 ngle number (likelihood) as a function of cosmological parameters. Thi
 s method provides uncertainty quantification via the full posterior of
  cosmological parameters\, which paves the way for a complete and opti
 mal cosmological data analysis with Normalizing Flows.
URL:https://www.physics.wisc.edu/events/?id=6527
END:VEVENT
END:VCALENDAR