BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Cosmology as an Optimisation Problem - T. Lucas Makinen (DAMTP)
DTSTART:20251110T160000Z
DTEND:20251110T170000Z
UID:TALK239770@talks.cam.ac.uk
CONTACT:65128
DESCRIPTION:Much of modern cosmology can be thought of as a multi-stage op
 timisation problem\, with the core objective of 1) constraining model para
 meters from data and 2) understanding where that information comes from. S
 imulation-based inference (SBI) utilises AI to unify simulation and parame
 ter inference under one roof\, but does not always a) leverage or b) excee
 d human domain knowledge in physical inference problems in terms of bits e
 xtracted from data.\n\nI will present an information-theoretic approach to
  illustrate SBI\, which can be naturally extended to derive *hybrid statis
 tics*\, an optimal framework for combining domain knowledge and learned ne
 ural summaries. These statistics improve information extraction compared t
 o neural summaries alone or their concatenation to existing summaries and 
 makes inference robust in settings with low training data. \n\nWe will sho
 w an application to DES Y3 weak lensing mock simulations\, forecast to ext
 ract a factor of 2 more information about the dark energy equation of stat
 e than existing traditional and neural methods. We will discuss how the mo
 dular nature of hybrid statistics might shed light on where non-Gaussian s
 ignatures of Dark Energy information might lie in weak lensing maps\, to b
 e exploited in upcoming Stage IV analyses.
LOCATION:Martin Ryle Seminar Room\, KICC
END:VEVENT
END:VCALENDAR