BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Multi-Scenario Failure Training for Robust Charged-Pion Energy Rec
 onstruction
DTSTART;VALUE=DATE-TIME:20261027T162500Z
DTEND;VALUE=DATE-TIME:20261027T163000Z
DTSTAMP;VALUE=DATE-TIME:20261101T051257Z
UID:indico-contribution-6399@cern.ch
DESCRIPTION:Speakers: Ruslan Yakunin (Kazan National Research Technologica
 l University)\nMachine-learning-based calorimeter reconstruction is usuall
 y optimized under fixed detector conditions\, whereas detector response ca
 n be affected by dead channels\, local readout failures\, gain drift\, and
  noise. We study whether exposing a reconstruction model to several simula
 ted failure modes during training improves its robustness compared with tr
 aining on a single corruption type. The study uses 533-voxel charged-pion 
 showers from CaloChallenge Dataset 1 and a layer-gated Transformer supplie
 d with voxel energies and operational masks. Two models with identical arc
 hitecture were trained: one exclusively on events with 5% randomly disable
 d cells\, and one on an equal mixture of clean events\, random cell loss\,
  contiguous dead-cell blocks\, and complete-layer failures. Evaluation was
  performed on an independent sample of 120\,800 showers under clean\, fami
 liar\, and unseen detector distortions. Multi-scenario training reduced re
 lative mean absolute error from 0.1804 to 0.1533 on clean events\, from 0.
 1799 to 0.1610 for block failures\, and from 0.1940 to 0.1751 for a mixed 
 unseen corruption. For physically consistent complete-layer loss\, uniform
  layer-averaged rMAE decreased from 0.2404 to 0.1929\, while voxel-weighte
 d rMAE decreased from 0.2844 to 0.2104. The gain was not universal: the si
 ngle-corruption regimen remained competitive near its specialized random-c
 ell-loss condition. These results show that multi-scenario failure trainin
 g improves broad robustness of charged-pion energy reconstruction while pr
 eserving a trade-off between generalization and narrow specialization.\n\n
 https://indico.particle.mephi.ru/event/626/contributions/6399/
LOCATION:Sevastopol hotel Conference Hall 1
URL:https://indico.particle.mephi.ru/event/626/contributions/6399/
END:VEVENT
END:VCALENDAR