BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Applying Machine Learning to the WASA-FRS experiment analysis
DTSTART;VALUE=DATE-TIME:20251120T090000Z
DTEND;VALUE=DATE-TIME:20251120T091500Z
DTSTAMP;VALUE=DATE-TIME:20260424T102353Z
UID:indico-contribution-29017@indico.ific.uv.es
DESCRIPTION:Speakers: David Calonge (CSIC)\nThe WASA-FRS HypHI Experiment 
 focuses on the study of light hypernuclei by means of heavy-ion induced\nr
 eactions in 6Li collisions with 12C at 1.96GeV/u. It is part of the WASA-F
 RS experimental campaign\, and so\nis the eta-prime experiment [1]. The di
 stinctive combination of the high-resolution spectrometer FRagment\nSepara
 tor (FRS) [2] and the high-acceptance detector system WASA [3] is used. Th
 e experiment was success-\nfully conducted at GSI-FAIR in Germany in March
  2022 as a component of the FAIR Phase-0 Physics Program\,\nwithin the Sup
 er-FRS Experiment Collaboration. The primary objectives of this experiment
  are twofold: to\nshed light on the hypertriton puzzle [4] and to investig
 ate the existence of the previously proposed nnΛ bound\nstate [5]. Curren
 tly\, the data from the experiment is under analysis.\nPart of the data an
 alysis is to provide a precise ion-optics of the measurement of the fragme
 nt originated from\nthe mesonic weak decay of the hypernuclei of interest.
  The reconstruction the ion-optics of fragments is based\non the calibrati
 on run of FRS optics. We have proposed to implement machine learning model
 s and neural\nnetworks to represent the ion-optics of FRS: While the curre
 nt state of the problem involves solving equations\nof motion of particles
  in non-ideal magnetic fields - which leads to the application of approxim
 ations in the\ncalculations - the implementation of data-driven models all
 ows us to obtain accurate results with possible\nbetter momentum and angul
 ar resolution.\nAnother important contribution to the analysis would be th
 e correct identification of signal versus background\nin the experimental 
 data. For this purpose\, we present an analysis using ML techniques as opp
 osed to typical\nselection conditions methods. The interest of this new ap
 proach comes from the fact that the models interpret\nthe physics behind t
 he data by making more accurate cuts and more consistent with the experime
 nt.\nIn this presentation\, we will show two different results of the curr
 ent status of the R&D in machine learning\nmodel of the ion-optics and the
  prospect of the inference of the track parameters of the fragments based 
 on the\ncalibration data recorded during the WASA-FRS experimental campaig
 n of 2022 and the signal to background\nratio enhancement with ML. For the
  ion optics part: our model selection optimization follows this approach:\
 nwe utilize AutoML environments [6]\, to determine the best pipeline for o
 ur data. Once identified\, this opti-\nmized pipeline is implemented in a 
 PyTorch model. Regarding the signal to background ratio enhancement\,\nwe 
 will make use of autoML libraries such as autogluon [7] to identify the H3
 Λ hypernuclei present in the\nexperimental datafile.\nThe results of this
  study demonstrate a robust reconstruction of the track angles in the FRS 
 mid-focal plane\,\nachieving an improvement of up to a ~40%. A resolution 
 of 0.65 mrad and 0.46 mrad was achieved for the\nhorizontal and vertical a
 ngular track plane\, respectively. Additionally\, the reconstruction of th
 e magnetic\nrigidity in the final focal plane attained a resolution Δp/p 
 of 5 10⁻⁴. From these results\, we demonstrated that\na data-driven mo
 del of non-linear ion optics is feasible. We also observed that training t
 he full model can be\nachieved very quickly\, paving the way for online tr
 aining during data collection at the FRS. This capability\nwill enable mor
 e accurate real-time analysis of fragment identification and improve the q
 uality of the exotic\nbeam obtained from the fragment separator.\nAlso\, a
  correct identification of signal events in the experimental data has also
  been carried out\, which al-\nlows a precise analysis of the properties o
 f the H3Λ from the experimental data\, such as the lifetime of the\nhyper
 nuclei.\n[1] Y.K. Tanaka et al.\, J. Phys. Conf. Ser. 1643 (2020) 012181.\
 n[2] H. Geissel et al.\, Nucl. Instr. and Meth. B 70 (1992) 286-297.\n[3] 
 C. Bargholtz et al.\, Nucl. Instr. and Meth. A 594 (2008) 339-350.\n[4] T.
 R. Saito et al.\, Nature Reviews Physics 3 (2021) 803-813.\n[5] C. Rappold
  et al.\, Phys. Rev. C 88 (2013) 041001.\n[6] M. Feurer et al.\, JMLR 23 2
 61 (2022) 1-61.\n[7] N. Erickso et al.\, 7th ICML Workshop on AutoML (2020
 ).\n\nhttps://indico.ific.uv.es/event/8035/contributions/29017/
LOCATION:
URL:https://indico.ific.uv.es/event/8035/contributions/29017/
END:VEVENT
END:VCALENDAR