Istituto Nazionale di Fisica Nucleare
Sezione di Firenze

DeepHep-ino. AI computing technologies for HEP

La versione in italiano segue. 

deephep ino logo

 

DeepHep-ino is an INFN initiative aiming at the investigation of the software technologies developed in the area of Deep Learning and Artificial Intelligence to the problems that the community of High-Energy Physics is facing.

The novel computing technologies developed to effectively exploit the High Performance Computing resources will play a crucial role in the future of High Energy Physics. In order to increase the statistical precision of the measurements performed with the large experiments of the LHC, more and more complicated algorithms are executed in real-time in order to obtain extremely pure samples. Both the real-time reconstruction of the collision events and the physical interpretation of the huge and pure selected samples require huge computing power that can benefit of the High Performance Computing resources such as GPU and other hardware accelerators.

The LHCb experiment at the LHC is a golden example of the paradigmatic change in the computing for High-Energy Physics. With the major upgrade of its detector and data-processing strategy that is being commissioned, LHCb will acquire and reconstruct in real-time 30 million collision events per second storing on disk only the reconstructed object needed for the statistical interpretation of the datasets and the achievement of the physics measurements.

DeepHep-ino focusses on LHCb as case of study to investigate the application of software libraries such as TensorFlow and Torch to High-Energy Physics.

To enhance the contributions from the Data Science and Artificial Intelligence communities, INFN funds brief internships at CERN for undergraduate students in Physics, Informatics and Software Engineering to develop and deploy algorithms with modern computing technologies to open problems in High Energy Physics.

Further readings

  • The LHCb experiment at CERN, lhcb.cern.ch
  • Technical Design Report for the upgrade of the LHCb Computing Model, cern.ch/go/7STL
  • Proposed activities for undergraduate students: cern.ch/go/f87Z
  • Contact person: Questo indirizzo email è protetto dagli spambots. È necessario abilitare JavaScript per vederlo. 

 

 

DeepHep-ino è un’iniziativa INFN che ha l’obiettivo di investigare le tecnologie software sviluppate nell’ambito dell’Intelligenza Artificiale e del Deep Learning per risolvere i problemi che i fisici delle alte energie stanno affrontando nella preparazione dei futuri esperimenti.

 

Le nuove tecnologie computazionali sviluppate per sfruttare in modo efficace le risorse the calcolo ad alte prestazioni giocherà un ruolo cruciale nel future della fisica delle alte energie. Al fine di accrescere la precision statistica delle misure effettuate con i grandi esperimenti di LHC, algoritmi sempre più complessi sono eseguiti in tempo-reale per ottenere campioni purissimi. Sia la ricostruzione in tempo reale che l’interpretazione statistica e fisica dei campioni sempre più grandi e puri richiede un’enorme potenza di calcolo che potrebbe beneficiare delle risorse di calcolo ad alte prestazioni quali GPU ed altri acceleratori hardware.

L’esperimento LHCb ad LHC è un ottimo esempio del cambio di paradigma in atto nel calcolo per gli esperimenti di fisica delle alte energie. Con l’upgrade del suo rivelatore della strategia di processamento dei dati che sono in corso d’installazione, LHCb acquisirà e ricostruirà in tempo-reale 30 millioni di collisioni protone-protone ogni secondo, salvado su disco solo gli elementi ricostruiti essenziali all’interpretazione statistica ed al successo delle misure fisiche.

DeepHep-ino è focalizzato su LHCb come caso di studio per per l’applicazione di librerie software quali TensorFlow e Torch alla fisica delle alte energie.

Al fine di aumentare il contatto della comunità della fisica delle alte energie con quelle della scienza dei dati e dell’intelligenza artificiale, l’Istituto Nazionale di Fisica Nucleare finanzia brevi tirocini al CERN di student in Fisica, Informatica ed Ingegneria Informatica per sviluppare ed applicare algoritmi con tecnologie computazionali moderne a problemi aperti nella comunità della fisica delle alte energie.

Approfondimenti

  • L’esperimento LHCb al CERN, lhcb.cern.ch
  • Progetto per l’upgrade del modello di calcolo di LHCb, cern.ch/go/7STL
  • Attività proposte per gli student magistrali, cern.ch/go/f87Z
  • Referente: Questo indirizzo email è protetto dagli spambots. È necessario abilitare JavaScript per vederlo.

 Presentazioni a conferenze

  • ACAT2019, Saas Fee, Switzerland, "Fast data-driven simulation of Cherenkov detectors using generative adversarial neural networks" [link]
  • CHEP2019, Adelaide, Australia, "Fast Simulations at LHCb" [link]
  • NeurlPS - ML4PS, Vancouver, Canada, "Data Driven Simulation of Cherenkov Detectors using Generative Adversarial Network" [link]

Altri documenti

  • "Performance of the Lamarr Prototype: the ultra-fast simulation option integrated in the LHCb simulation framework", LHCb public Figure 2019-017 [link]

PortaleINFN meetings of the week ita
bottone sol WebMail

Calendario Eventi

Aprile 2024
L M M G V S D
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21
22 23 24 25 26 27 28
29 30 1 2 3 4 5

prossimi appuntamenti

Seminari Fisica Teorica
16 Apr 2024    02:30PM - 03:30PM
Snipe Hunt: using the largest cavities on earth to search for DM
Seminari Fisica Teorica
17 Apr 2024    02:30PM - 03:30PM
Entanglement in interacting Majorana chains and transitions of von Neumann algebras
Seminari Fisica Teorica
24 Apr 2024    02:30PM - 03:30PM
TBA
PortaleINFN meetings of the week ita
bottone sol WebMail

Calendario Eventi

Aprile 2024
L M M G V S D
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21
22 23 24 25 26 27 28
29 30 1 2 3 4 5

prossimi appuntamenti

Seminari Fisica Teorica
16 Apr 2024    02:30PM - 03:30PM
Snipe Hunt: using the largest cavities on earth to search for DM
Seminari Fisica Teorica
17 Apr 2024    02:30PM - 03:30PM
Entanglement in interacting Majorana chains and transitions of von Neumann algebras
Seminari Fisica Teorica
24 Apr 2024    02:30PM - 03:30PM
TBA

Please publish modules in offcanvas position.