BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Building Reproducible Machine Learning Pipelines for inference of 
 Galaxy Properties at Scale  - Gurjeet Jagwani - IoA\, University of Cambri
 dge
DTSTART:20251023T120000Z
DTEND:20251023T130000Z
UID:TALK234391@talks.cam.ac.uk
CONTACT:Jack Atkinson
DESCRIPTION:Next-generation astronomical surveys like the Legacy Survey of
  Space and Time (LSST) will deliver billions of galaxy observations crucia
 l for understanding dark matter and dark energy. However\, extracting reli
 able galaxy properties like redshifts from these data requires scalable co
 mputational approaches that can handle these enormous datasets while maint
 aining scientific rigour and reproducibility. We present pop-cosmos\, a fo
 rward-modelling framework for photometric galaxy survey data that constrai
 ns population-level galaxy properties up to redshift 6. Galaxies are model
 led as draws from a population prior over physical parameters (redshift\, 
 stellar mass\, star formation history\, dust properties)\, mapped to obser
 ved colors and brightness using neural emulators of complex astrophysical 
 models—achieving 10000x speedups. We use simulation-based inference to c
 alibrate this population prior on deep multi-wavelength data (COSMOS2020)\
 , training a diffusion model to match the statistical properties of real s
 urvey data. The resulting model helps us understand and probe various astr
 ophysical and cosmological phenomena. Central to our framework is flowfusi
 on\, a general-purpose library for density estimation and generative model
 ling that implements state-of-the-art machine learning methods including d
 iffusion models and flow-matching. I will demonstrate how our open-source 
 toolkit enables reproducible results from our scientific applications and 
 discuss ongoing work with the Kilo-Degree Survey in preparation for LSST.
LOCATION:Room E\, West Hub
END:VEVENT
END:VCALENDAR