BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Heidelberg University//HePhySTO//EN
CALSCALE:GREGORIAN
BEGIN:VEVENT
DTEND:20260611T101500Z
UID:41d45488b819878678d279e0771bff0f@physik.uni-heidelberg.de
DTSTAMP:20260411T180743Z
LOCATION:ARI\, Moenchhofstrasse 12-14\, Seminarraum 1.OG
DESCRIPTION:Interacting and merging galaxies are crucial for the understand
 ing of how galaxies assemble and\nevolve\, moreover they also provide a uni
 que probe of cosmological models of the universe. However\,\nclassifying ga
 laxies as interacting remains a major challenge. With the Q1 release of Euc
 lid a new dataset\nof high-quality photometric and spectroscopic data is av
 ailable\, enabling the assembly of a large statistical sample of galaxies u
 ndergoing mergers. To create such a sample\, we make use of a novel semi-su
 pervised machine learning tool dubbed "AnomalyMatch"\, designed to identify
  interesting objects. AnomalyMatch requires only a small number of initial 
 labelled training samples (< 100) and allows one to find objects of interes
 t quickly via user-in-the-loop active learning. The output of the machine l
 earning model is used to model the distribution of mergers\, relying on Bay
 esian analysis and Markov Chain Monte Carlo on the basis of expert labelled
  images.\nTo determine its performance\, the model and Bayesian Inference i
 s evaluated on synthetic imagery stemming from the IllustrisTNG simulation 
 where it manages to reproduce the underlying fraction of mergers. The appli
 cation to a mass-complete sample of about 913k Euclid galaxies with 0.2 < z
  < 1.0\, indicates a decreasing merger fraction evolution ~0.2 over the ent
 ire mass range\, \nwhich is in contrast to some of the literature. \nAdopti
 ng recent estimates for observational timescales\, we compare the fractiona
 l galaxy merger rate to expectations from the Illustris simulations and fin
 d that our results indicate increasing\, but higher absolute galaxy merger 
 rates than previous work. Assuming a certain distribution of pre and post-m
 ergers in our sample recover absolute values\, but flatter evolution than t
 he Illustris simulation predicts.\n\nImported from https://www.physik.uni-h
 eidelberg.de/hephysto/ (no warranty for accuracy).
URL;VALUE=URI:https://www.physik.uni-heidelberg.de/hephysto/index.php?s=tal
 k&id=12570
SUMMARY:ARI Institute Colloquium: Laslo Ruhberg - Investigating Galaxy-Merg
 er Rates in Euclid Q1 with Semi-Supervised Machine Learning
DTSTART:20260611T091500Z
END:VEVENT
END:VCALENDAR