BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Physics behind Transformer jet classification
DTSTART:20251117T130000Z
DTEND:20251117T150000Z
DTSTAMP:20260827T095200Z
UID:indico-event-1027@indico.hiskp.uni-bonn.de
DESCRIPTION:Speakers: Mihoko Nojiri\n\nDeep learning has achieved unpreced
 ented performance in particle physics. In particular\, the use of transfo
 rmers has shown great promise for jet identification.\nOn the other hand\,
  considering that jets originate from QCD dynamics\, a transformer archite
 cture designed to incorporate relevant physical inductive biases may repro
 duce the performance of general-purpose transformers with a significantly 
 smaller number of model parameters.\nFrom our research\, we introduce seve
 ral ideas along these lines\, such as the use of cross-attention between s
 ubjets (jets) and constituent particles within a jet\, as well as attentio
 n matrices restricted to information from pair wise variables. \n\nhttps:
 //indico.hiskp.uni-bonn.de/event/1027/
URL:https://indico.hiskp.uni-bonn.de/event/1027/
END:VEVENT
END:VCALENDAR