BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20210518T152206Z
LOCATION:Booth N-230
DTSTART;TZID=Europe/Stockholm:20190618T151500
DTEND;TZID=Europe/Stockholm:20190618T154500
UID:isc_hpc_ISC High Performance 2019_sess208_proj137@linklings.com
SUMMARY:(PP28): Evolutionary Convolutional Neural Network for High Energy 
 Physics Detector Simulation
DESCRIPTION:Project Poster\n\n(PP28): Evolutionary Convolutional Neural Ne
 twork for High Energy Physics Detector Simulation\n\nKhattak, Vallecorsa, 
 Carminati, Khan\n\nHigh Energy Physics relies on Monte Carlo (MC) for diff
 erent aspects of data analysis. MC simulation implement complex computatio
 ns that, today, result in ~50% of CERN Computing Grid resources. Several a
 lternative approaches are being investigated trading some accuracy for spe
 ed. Deep Learning approaches, for example, can result in x3000 speed-up wh
 ile retaining 10% accuracy with respect to MC.\n\nDesign and optimisation 
 of neural networks is very difficult and time consuming. Evolutionary comp
 uting could be exploited to improve this process, while providing the bene
 fit of both network training and topology optimization in one step. By cod
 ing a neural network into genes and chromosome, Genetic Algorithms (GA) wo
 uld exhibit numerous advantages: easily parallelizable, they could help re
 ach global instead of local optima often selected by Stochastic Gradient D
 escent (SGD); they could also be combined to SGD for fine tuning.  \n\nOur
  project intends to use GA to train and optimise a 3-dimensional Convoluti
 onal Generative Adversarial Network we have developed for particle detecto
 r simulation. It is implemented in a few steps: a first phase, will implem
 ent a simplified version of our GAN network, focusing on one of the two ne
 tworks that constitute the GAN (the “discriminator”). We will design a fir
 st prototype and define the method applicability (in terms of network comp
 lexity and computing resources). We will gradually increase the complexity
  of the problem, in terms of input image size and network features and fin
 ally implement the full adversarial training approach. We will compare per
 formances to standard SGD and classical optimisation approaches (i.e. Baye
 sian).\n\nPasses: Conference Pass, Exhibition Pass\n\nTag: AI/Machine Lear
 ning/Deep Learning, Networks, Visualization & Virtual Reality
URL:https://2019.isc-program.com/presentation/?id=proj137&sess=sess208
END:VEVENT
END:VCALENDAR

