Generic searches in ATLAS ITN: MVA for New Physic 27/04/2016 - PowerPoint PPT Presentation

Generic searches in ATLAS ITN: MVA for New Physic 27/04/2016 – présentations thèses Julien Donini

ITN project: AMVA4NewPhysics “Advanced Multivariate analysis for New Physics searches” Innovative Training Network (ITN): EU Marie Curie Actions H2020 Aimed at training (PhD) students and performing research Develop and improve advanced stat learning tools for data analysis Training of students in stat learning, computational tools, data science Perform studies in Higgs sector and searches for new physics at LHC  Network of 8 institutes and 6 partners (4 non-academic)  Officially started 1 September 2015 , duration: 48 Months  Recruitment a total of 10 ESR (“Early Stage Researchers)  Recruitment of 1 ESR at LPC by September: see Call for PhD  ATLAS-LPC contributors: JD (PI), D. Calvet, E. Busato Julien Donini

→ Coordinator: LPC Julien Donini

Scientific goals of WP2 Study LHC Run 2 data for new physics searches Specialize/Optimize Statistical Learning methods Comparison study of advanced MVA algorithms  Signature-specific new physics searches (CMS)  Global searches considering wide number of signatures (ATLAS)  Develop tools for reinterpretation of MVA results for NP searches Main beneficiary of this WP 4 ESR will work on WP2 UBP, IASA, CERN, TUM  UBP (ATLAS) : 36m Other contributors  IASA (CMS) : 36m UNIPD, INFN, UCL, UCI, EPFL  CERN (TH): 36m Partners (secondments)  CERN (CMS): 24m Yandex, MW, SDG List of all contributors : http://www.pd.infn.it/AMVA4NewPhysics/allparticipants.html Julien Donini 4

General work strategy and interactions Main beneficiary Test algorithms for simultaneous investigation of multi-dimensional  datasets (UBP) Test MVA algorithms on specific NP signatures, maximize search  sensitivity (IASA, CERN) Develop ATOM package to reinterpret MVA searches for any model  (CERN/TUM) Interaction with other contributors to WP2: 1. Use of the MEM (WP3: UCL, CERN) 2. Implement WP4 developments by INFN, UNIPD, and UCI, • Ex: UCI expertise on deep neural networks 3. Development of different semi-supervised technique (EPFL) 4. Global classification tools (guidance from UNIPD) Julien Donini 5

Signature-specific searches Example: search for W’ bosons decaying to a top and a b quark → appear in many extensions of the SM: L/R models, KK excitations … Mass range: 0.5 – 3.0 TeV Four BDT to separate S/B Limits on signal cross-section Limits on effective couplings Physics Letters B 743 (2015) 235-255 Julien Donini 6

Global searches Example: general search for new phenomena with ATLAS Topologies involving lepton, photons, jets and missing transverse momentum. Kinematic distributions are scanned for deviations from the SM prediction. If a significant deviation is found, a dedicated analysis will be required. ATLAS-CONF-2014-006 Julien Donini 7

backup Julien Donini 8

Model testing: ATOM (Weiler et al.) Replicates exp. analyses and automatically checks if a model is excluded. Test of beyond the standard model versus existing data. www.research.kobe-u.ac.jp/fsci-pacos/seminar_files/20140528_sakurai.pdf Julien Donini 9

Gantt chart Julien Donini 10