BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:χ² and Beyond: Robust Statistics for Hadronisation Tuning
DTSTART:20261113T130000Z
DTEND:20261113T140000Z
DTSTAMP:20261101T061400Z
UID:indico-event-19109@indico.global
DESCRIPTION:Speakers: James Whitehead\n\nMonte Carlo Event Generators in u
 se for LHC phenomenology combine perturbative modelling of hard processes 
 with semi-empirical models of non-perturbative physics\, including hadroni
 sation and the underlying event. These models introduce parameters that mu
 st be calibrated\, or 'tuned'\, to data. Standard tuning approaches minimi
 se a weighted χ² objective\, so large residuals can exert disproportiona
 te influence. Yet such discrepancies are expected to arise from imperfect 
 phenomenological models\, and may reflect structural limitations that cann
 ot be resolved by adjusting the parameters. Consequently\, a tune can be d
 riven by precisely the measurements which the model is least able to descr
 ibe.\nThe concept of ‘robustness’\, developed in the statistical frame
 work of Huber and others\, offers a principled way to limit this sensitivi
 ty. Robust statistics concerns the stability of inference under departures
  from assumed statistical distributions\, including contamination by obser
 vations drawn from a different distribution. We compare χ² with robust a
 lternatives\, both through loss-function minimisation and Bayesian inferen
 ce with MCMC sampling. We apply these approaches to the final stage of a r
 ecent Herwig 7 tune using the Pythia Lund string model\, and assess how th
 ese statistical choices affect the inferred parameters\, their uncertainti
 es\, and sensitivity to tuning weights.\n \n\nhttps://indico.global/event
 /19109/
LOCATION:Bohr Conference Room
URL:https://indico.global/event/19109/
END:VEVENT
END:VCALENDAR