BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:2
UID:UW-Physics-Event-6244
DTSTART:20210210T170000Z
DTEND:20210210T181500Z
DTSTAMP:20260308T231933Z
LAST-MODIFIED:20210206T184340Z
LOCATION:Online Seminar: Please sign up for our mailing list at www.ph
 ysicsmeetsml.org for zoom link
SUMMARY:Physics meets ML to solve cosmological inference\, Physics ∩
  ML Seminar\, Ben Wandelt\, Institut d’Astrophysique de Paris / Inst
 itut Lagrange\, Sorbonne University and Center for Computational Astro
 physics\, Flatiron Institute\, New York
DESCRIPTION:The goal of cosmological inference is to learn about the o
 rigin\, composition\, evolution\, and fate of the cosmos from all acce
 ssible sources of astronomical data\, such as the cosmic microwave bac
 kground\, galaxy surveys\, or electromagnetic and gravitational wave t
 ransients. Traditionally\, the field has progressed by designing and m
 odeling intuitive summaries of the data\, such as n-point correlations
 . This traditional approach has a number of risks and limitations: how
  do we know if we computed the most informative statistics? Did we for
 get any summaries that would have provided additional information or b
 reak parameter degeneracies? Did we take into account all the ways the
  model is affecting the data? To be feasible\, the traditional approac
 h imposes approximations on the statistical modeling (e.g. the likelih
 ood form) and on the physical modeling. I will discuss a new mode of c
 osmological inference: simulation-based\, full-physics modeling\, made
  feasible through multiple advances in 1) machine-learning\, 2) in the
  way we design and run simulations of cosmological observables\, and 3
 ) in how we compare models to data. The goal is to use current and nex
 t generation data to reconstruct the cosmological initial conditions a
 nd constrain cosmological physics much more completely than has been f
 easible in the past. I will discuss current status\, and ways to meet 
 the new challenges inherent in this approach\, including robustness to
  model misspecification.
URL:https://www.physics.wisc.edu/events/?id=6244
END:VEVENT
END:VCALENDAR